id sid tid token lemma pos fbem-1515 1 1 frontiers frontier NOUN fbem-1515 1 2 in in ADP fbem-1515 1 3 business business NOUN fbem-1515 1 4 , , PUNCT fbem-1515 1 5 economics economic NOUN fbem-1515 1 6 and and CCONJ fbem-1515 1 7 management management NOUN fbem-1515 1 8 issn issn PROPN fbem-1515 1 9 : : PUNCT fbem-1515 1 10 2766 2766 NUM fbem-1515 1 11 - - PUNCT fbem-1515 1 12 824x 824x NUM fbem-1515 1 13 | | ADJ fbem-1515 1 14 vol vol NOUN fbem-1515 1 15 . . PROPN fbem-1515 2 1 5 5 NUM fbem-1515 2 2 , , PUNCT fbem-1515 2 3 no no INTJ fbem-1515 2 4 . . NOUN fbem-1515 2 5 1 1 NUM fbem-1515 2 6 , , PUNCT fbem-1515 2 7 2022 2022 NUM fbem-1515 2 8 142 142 NUM fbem-1515 2 9 prediction prediction NOUN fbem-1515 2 10 of of ADP fbem-1515 2 11 car car NOUN fbem-1515 2 12 loan loan NOUN fbem-1515 2 13 default default NOUN fbem-1515 2 14 results result NOUN fbem-1515 2 15 based base VERB fbem-1515 2 16 on on ADP fbem-1515 2 17 multi multi ADJ fbem-1515 2 18 model model NOUN fbem-1515 2 19 fusion fusion NOUN fbem-1515 2 20 zhaoyang zhaoyang PROPN fbem-1515 2 21 hong hong PROPN fbem-1515 2 22 , , PUNCT fbem-1515 2 23 wenxuan wenxuan PROPN fbem-1515 2 24 deng deng PROPN fbem-1515 2 25 , , PUNCT fbem-1515 2 26 xiuyu xiuyu PROPN fbem-1515 2 27 gong gong VERB fbem-1515 2 28 the the DET fbem-1515 2 29 hong hong PROPN fbem-1515 2 30 kong kong PROPN fbem-1515 2 31 university university PROPN fbem-1515 2 32 of of ADP fbem-1515 2 33 science science NOUN fbem-1515 2 34 and and CCONJ fbem-1515 2 35 technology technology NOUN fbem-1515 2 36 , , PUNCT fbem-1515 2 37 kowloon kowloon PROPN fbem-1515 2 38 , , PUNCT fbem-1515 2 39 hong hong PROPN fbem-1515 2 40 kong kong PROPN fbem-1515 2 41 , , PUNCT fbem-1515 2 42 china china PROPN fbem-1515 2 43 abstract abstract PROPN fbem-1515 2 44 : : PUNCT fbem-1515 2 45 with with SCONJ fbem-1515 2 46 the the DET fbem-1515 2 47 prosperity prosperity NOUN fbem-1515 2 48 and and CCONJ fbem-1515 2 49 development development NOUN fbem-1515 2 50 of of ADP fbem-1515 2 51 the the DET fbem-1515 2 52 asset asset NOUN fbem-1515 2 53 management management NOUN fbem-1515 2 54 industry industry NOUN fbem-1515 2 55 and and CCONJ fbem-1515 2 56 various various ADJ fbem-1515 2 57 financial financial ADJ fbem-1515 2 58 derivatives derivative NOUN fbem-1515 2 59 , , PUNCT fbem-1515 2 60 many many ADJ fbem-1515 2 61 micro micro NOUN fbem-1515 2 62 - - NOUN fbem-1515 2 63 loans loan NOUN fbem-1515 2 64 and and CCONJ fbem-1515 2 65 online online ADJ fbem-1515 2 66 loans loan NOUN fbem-1515 2 67 have have AUX fbem-1515 2 68 gradually gradually ADV fbem-1515 2 69 entered enter VERB fbem-1515 2 70 the the DET fbem-1515 2 71 public public ADJ fbem-1515 2 72 view view NOUN fbem-1515 2 73 . . PUNCT fbem-1515 3 1 how how SCONJ fbem-1515 3 2 to to PART fbem-1515 3 3 predict predict VERB fbem-1515 3 4 the the DET fbem-1515 3 5 default default NOUN fbem-1515 3 6 probability probability NOUN fbem-1515 3 7 of of ADP fbem-1515 3 8 customer customer NOUN fbem-1515 3 9 loans loan NOUN fbem-1515 3 10 is be AUX fbem-1515 3 11 a a DET fbem-1515 3 12 hot hot ADJ fbem-1515 3 13 topic topic NOUN fbem-1515 3 14 in in ADP fbem-1515 3 15 the the DET fbem-1515 3 16 market market NOUN fbem-1515 3 17 . . PUNCT fbem-1515 4 1 therefore therefore ADV fbem-1515 4 2 , , PUNCT fbem-1515 4 3 in in ADP fbem-1515 4 4 this this DET fbem-1515 4 5 paper paper NOUN fbem-1515 4 6 , , PUNCT fbem-1515 4 7 by by ADP fbem-1515 4 8 collecting collect VERB fbem-1515 4 9 the the DET fbem-1515 4 10 data data NOUN fbem-1515 4 11 profile profile NOUN fbem-1515 4 12 of of ADP fbem-1515 4 13 more more ADJ fbem-1515 4 14 than than ADP fbem-1515 4 15 10 10 NUM fbem-1515 4 16 thousand thousand NUM fbem-1515 4 17 car car NOUN fbem-1515 4 18 loan loan NOUN fbem-1515 4 19 borrowers borrower NOUN fbem-1515 4 20 and and CCONJ fbem-1515 4 21 fitting fit VERB fbem-1515 4 22 the the DET fbem-1515 4 23 fusion fusion NOUN fbem-1515 4 24 model model NOUN fbem-1515 4 25 of of ADP fbem-1515 4 26 4 4 NUM fbem-1515 4 27 methods method NOUN fbem-1515 4 28 : : PUNCT fbem-1515 4 29 logistic logistic ADJ fbem-1515 4 30 model model NOUN fbem-1515 4 31 , , PUNCT fbem-1515 4 32 decision decision NOUN fbem-1515 4 33 model model NOUN fbem-1515 4 34 , , PUNCT fbem-1515 4 35 random random ADJ fbem-1515 4 36 forest forest NOUN fbem-1515 4 37 , , PUNCT fbem-1515 4 38 and and CCONJ fbem-1515 4 39 knn knn PROPN fbem-1515 4 40 model model NOUN fbem-1515 4 41 to to ADP fbem-1515 4 42 the the DET fbem-1515 4 43 data datum NOUN fbem-1515 4 44 , , PUNCT fbem-1515 4 45 the the DET fbem-1515 4 46 author author NOUN fbem-1515 4 47 examines examine VERB fbem-1515 4 48 the the DET fbem-1515 4 49 behavioral behavioral ADJ fbem-1515 4 50 data datum NOUN fbem-1515 4 51 of of ADP fbem-1515 4 52 borrowers borrower NOUN fbem-1515 4 53 to to PART fbem-1515 4 54 predict predict VERB fbem-1515 4 55 whether whether SCONJ fbem-1515 4 56 the the DET fbem-1515 4 57 borrowers borrower NOUN fbem-1515 4 58 will will AUX fbem-1515 4 59 default default VERB fbem-1515 4 60 in in ADP fbem-1515 4 61 the the DET fbem-1515 4 62 future future NOUN fbem-1515 4 63 and and CCONJ fbem-1515 4 64 find find VERB fbem-1515 4 65 the the DET fbem-1515 4 66 best good ADJ fbem-1515 4 67 threshold threshold NOUN fbem-1515 4 68 to to PART fbem-1515 4 69 reach reach VERB fbem-1515 4 70 the the DET fbem-1515 4 71 lowest low ADJ fbem-1515 4 72 cost cost NOUN fbem-1515 4 73 . . PUNCT fbem-1515 5 1 the the DET fbem-1515 5 2 findings finding NOUN fbem-1515 5 3 indicate indicate VERB fbem-1515 5 4 that that SCONJ fbem-1515 5 5 our our PRON fbem-1515 5 6 final final ADJ fbem-1515 5 7 prediction prediction NOUN fbem-1515 5 8 can can AUX fbem-1515 5 9 reduce reduce VERB fbem-1515 5 10 costs cost NOUN fbem-1515 5 11 by by ADP fbem-1515 5 12 38.9 38.9 NUM fbem-1515 5 13 % % NOUN fbem-1515 5 14 . . PUNCT fbem-1515 6 1 the the DET fbem-1515 6 2 excellent excellent ADJ fbem-1515 6 3 result result NOUN fbem-1515 6 4 shows show VERB fbem-1515 6 5 that that SCONJ fbem-1515 6 6 this this DET fbem-1515 6 7 model model NOUN fbem-1515 6 8 can can AUX fbem-1515 6 9 be be AUX fbem-1515 6 10 applied apply VERB fbem-1515 6 11 to to ADP fbem-1515 6 12 the the DET fbem-1515 6 13 real real ADJ fbem-1515 6 14 market market NOUN fbem-1515 6 15 to to PART fbem-1515 6 16 help help VERB fbem-1515 6 17 lending lending NOUN fbem-1515 6 18 institutions institution NOUN fbem-1515 6 19 predict predict VERB fbem-1515 6 20 default default NOUN fbem-1515 6 21 results result NOUN fbem-1515 6 22 and and CCONJ fbem-1515 6 23 formulate formulate ADJ fbem-1515 6 24 strategies strategy NOUN fbem-1515 6 25 to to PART fbem-1515 6 26 avoid avoid VERB fbem-1515 6 27 default default NOUN fbem-1515 6 28 risks risk NOUN fbem-1515 6 29 in in ADP fbem-1515 6 30 the the DET fbem-1515 6 31 process process NOUN fbem-1515 6 32 of of ADP fbem-1515 6 33 borrowers borrower NOUN fbem-1515 6 34 ' ' PART fbem-1515 6 35 evaluation evaluation NOUN fbem-1515 6 36 according accord VERB fbem-1515 6 37 to to ADP fbem-1515 6 38 the the DET fbem-1515 6 39 model model NOUN fbem-1515 6 40 coefficient coefficient NOUN fbem-1515 6 41 . . PUNCT fbem-1515 7 1 keywords keyword NOUN fbem-1515 7 2 : : PUNCT fbem-1515 7 3 loan loan NOUN fbem-1515 7 4 default default NOUN fbem-1515 7 5 , , PUNCT fbem-1515 7 6 multi multi ADJ fbem-1515 7 7 - - ADJ fbem-1515 7 8 model model ADJ fbem-1515 7 9 fusion fusion NOUN fbem-1515 7 10 , , PUNCT fbem-1515 7 11 credit credit NOUN fbem-1515 7 12 default default NOUN fbem-1515 7 13 , , PUNCT fbem-1515 7 14 credit credit NOUN fbem-1515 7 15 risk risk NOUN fbem-1515 7 16 , , PUNCT fbem-1515 7 17 machine machine NOUN fbem-1515 7 18 learning learning NOUN fbem-1515 7 19 . . PUNCT fbem-1515 8 1 1 1 X fbem-1515 8 2 . . X fbem-1515 8 3 literature literature NOUN fbem-1515 8 4 review review PROPN fbem-1515 8 5 with with ADP fbem-1515 8 6 the the DET fbem-1515 8 7 deep deep ADJ fbem-1515 8 8 development development NOUN fbem-1515 8 9 of of ADP fbem-1515 8 10 big big ADJ fbem-1515 8 11 data datum NOUN fbem-1515 8 12 and and CCONJ fbem-1515 8 13 programming programming NOUN fbem-1515 8 14 tools tool NOUN fbem-1515 8 15 , , PUNCT fbem-1515 8 16 classification classification NOUN fbem-1515 8 17 and and CCONJ fbem-1515 8 18 regression regression NOUN fbem-1515 8 19 models model NOUN fbem-1515 8 20 are be AUX fbem-1515 8 21 used use VERB fbem-1515 8 22 to to PART fbem-1515 8 23 predict predict VERB fbem-1515 8 24 credit credit NOUN fbem-1515 8 25 crisis crisis NOUN fbem-1515 9 1 [ [ X fbem-1515 9 2 1 1 NUM fbem-1515 9 3 ] ] PUNCT fbem-1515 9 4 . . PUNCT fbem-1515 10 1 galindo galindo PROPN fbem-1515 10 2 used use VERB fbem-1515 10 3 machine machine NOUN fbem-1515 10 4 learning learn VERB fbem-1515 10 5 to to PART fbem-1515 10 6 research research VERB fbem-1515 10 7 credit credit NOUN fbem-1515 10 8 risk risk NOUN fbem-1515 10 9 and and CCONJ fbem-1515 10 10 got get VERB fbem-1515 10 11 the the DET fbem-1515 10 12 conclusion conclusion NOUN fbem-1515 10 13 that that SCONJ fbem-1515 10 14 decision decision NOUN fbem-1515 10 15 tree tree NOUN fbem-1515 10 16 performed perform VERB fbem-1515 10 17 best well ADV fbem-1515 10 18 in in ADP fbem-1515 10 19 terms term NOUN fbem-1515 10 20 of of ADP fbem-1515 10 21 classification classification NOUN fbem-1515 10 22 results result NOUN fbem-1515 10 23 among among ADP fbem-1515 10 24 the the DET fbem-1515 10 25 decision decision NOUN fbem-1515 10 26 tree tree NOUN fbem-1515 10 27 , , PUNCT fbem-1515 10 28 neural neural ADJ fbem-1515 10 29 networks network NOUN fbem-1515 10 30 , , PUNCT fbem-1515 10 31 and and CCONJ fbem-1515 10 32 k k X fbem-1515 10 33 - - PUNCT fbem-1515 10 34 nearest near ADJ fbem-1515 10 35 neighbor neighbor NOUN fbem-1515 10 36 methods method NOUN fbem-1515 10 37 in in ADP fbem-1515 10 38 the the DET fbem-1515 10 39 process process NOUN fbem-1515 10 40 of of ADP fbem-1515 10 41 credit credit NOUN fbem-1515 10 42 default default NOUN fbem-1515 10 43 prediction prediction NOUN fbem-1515 10 44 [ [ X fbem-1515 10 45 2 2 NUM fbem-1515 10 46 ] ] PUNCT fbem-1515 10 47 . . PUNCT fbem-1515 11 1 malekipirbazari malekipirbazari PROPN fbem-1515 11 2 et et PROPN fbem-1515 11 3 al al PROPN fbem-1515 11 4 . . PROPN fbem-1515 11 5 conducted conduct VERB fbem-1515 11 6 an an DET fbem-1515 11 7 empirical empirical ADJ fbem-1515 11 8 study study NOUN fbem-1515 11 9 in in ADP fbem-1515 11 10 the the DET fbem-1515 11 11 lending lending NOUN fbem-1515 11 12 club club NOUN fbem-1515 11 13 dataset dataset VERB fbem-1515 11 14 and and CCONJ fbem-1515 11 15 concluded conclude VERB fbem-1515 11 16 that that SCONJ fbem-1515 11 17 in in ADP fbem-1515 11 18 the the DET fbem-1515 11 19 process process NOUN fbem-1515 11 20 of of ADP fbem-1515 11 21 predicting predict VERB fbem-1515 11 22 borrower borrower NOUN fbem-1515 11 23 defaults default NOUN fbem-1515 11 24 , , PUNCT fbem-1515 11 25 the the DET fbem-1515 11 26 random random ADJ fbem-1515 11 27 forest forest NOUN fbem-1515 11 28 model model NOUN fbem-1515 11 29 outperformed outperform VERB fbem-1515 11 30 fico fico PROPN fbem-1515 11 31 scores score NOUN fbem-1515 11 32 and and CCONJ fbem-1515 11 33 lending lending NOUN fbem-1515 11 34 club club NOUN fbem-1515 11 35 ’s ’s PART fbem-1515 11 36 own own ADJ fbem-1515 11 37 credit credit NOUN fbem-1515 11 38 rating rating NOUN fbem-1515 11 39 methodology methodology NOUN fbem-1515 12 1 [ [ X fbem-1515 12 2 3 3 NUM fbem-1515 12 3 ] ] PUNCT fbem-1515 12 4 . . PUNCT fbem-1515 13 1 some some DET fbem-1515 13 2 scholars scholar NOUN fbem-1515 13 3 have have AUX fbem-1515 13 4 improved improve VERB fbem-1515 13 5 loan loan NOUN fbem-1515 13 6 default default NOUN fbem-1515 13 7 prediction prediction NOUN fbem-1515 13 8 by by ADP fbem-1515 13 9 fusing fuse VERB fbem-1515 13 10 different different ADJ fbem-1515 13 11 models model NOUN fbem-1515 13 12 [ [ X fbem-1515 13 13 4,5,6 4,5,6 NUM fbem-1515 13 14 ] ] PUNCT fbem-1515 13 15 , , PUNCT fbem-1515 13 16 including include VERB fbem-1515 13 17 gbdt gbdt NOUN fbem-1515 13 18 , , PUNCT fbem-1515 13 19 decision decision NOUN fbem-1515 13 20 tree tree NOUN fbem-1515 13 21 , , PUNCT fbem-1515 13 22 xgboost xgboost ADV fbem-1515 13 23 , , PUNCT fbem-1515 13 24 and and CCONJ fbem-1515 13 25 other other ADJ fbem-1515 13 26 models model NOUN fbem-1515 13 27 . . PUNCT fbem-1515 14 1 all all PRON fbem-1515 14 2 of of ADP fbem-1515 14 3 the the DET fbem-1515 14 4 above above ADJ fbem-1515 14 5 previous previous ADJ fbem-1515 14 6 studies study NOUN fbem-1515 14 7 provide provide VERB fbem-1515 14 8 the the DET fbem-1515 14 9 basis basis NOUN fbem-1515 14 10 for for SCONJ fbem-1515 14 11 this this DET fbem-1515 14 12 paper paper NOUN fbem-1515 14 13 to to PART fbem-1515 14 14 integrate integrate VERB fbem-1515 14 15 four four NUM fbem-1515 14 16 models model NOUN fbem-1515 14 17 : : PUNCT fbem-1515 14 18 logistic logistic ADJ fbem-1515 14 19 model model NOUN fbem-1515 14 20 , , PUNCT fbem-1515 14 21 decision decision NOUN fbem-1515 14 22 model model NOUN fbem-1515 14 23 , , PUNCT fbem-1515 14 24 random random ADJ fbem-1515 14 25 forest forest NOUN fbem-1515 14 26 , , PUNCT fbem-1515 14 27 and and CCONJ fbem-1515 14 28 knn knn VERB fbem-1515 14 29 to to PART fbem-1515 14 30 predict predict VERB fbem-1515 14 31 the the DET fbem-1515 14 32 car car NOUN fbem-1515 14 33 loan loan NOUN fbem-1515 14 34 default default NOUN fbem-1515 14 35 probability probability NOUN fbem-1515 14 36 . . PUNCT fbem-1515 15 1 the the DET fbem-1515 15 2 results result NOUN fbem-1515 15 3 are be AUX fbem-1515 15 4 presented present VERB fbem-1515 15 5 in in ADP fbem-1515 15 6 the the DET fbem-1515 15 7 following follow VERB fbem-1515 15 8 sections section NOUN fbem-1515 15 9 . . PUNCT fbem-1515 16 1 related relate VERB fbem-1515 16 2 models model NOUN fbem-1515 16 3 and and CCONJ fbem-1515 16 4 methods method NOUN fbem-1515 16 5 : : PUNCT fbem-1515 16 6 random random ADJ fbem-1515 16 7 forest forest NOUN fbem-1515 16 8 is be AUX fbem-1515 16 9 a a DET fbem-1515 16 10 classifier classifier NOUN fbem-1515 16 11 that that PRON fbem-1515 16 12 uses use VERB fbem-1515 16 13 multiple multiple ADJ fbem-1515 16 14 decision decision NOUN fbem-1515 16 15 trees tree NOUN fbem-1515 16 16 to to PART fbem-1515 16 17 train train VERB fbem-1515 16 18 and and CCONJ fbem-1515 16 19 predict predict VERB fbem-1515 16 20 samples sample NOUN fbem-1515 16 21 . . PUNCT fbem-1515 17 1 in in ADP fbem-1515 17 2 the the DET fbem-1515 17 3 process process NOUN fbem-1515 17 4 of of ADP fbem-1515 17 5 classification classification NOUN fbem-1515 17 6 , , PUNCT fbem-1515 17 7 for for ADP fbem-1515 17 8 each each DET fbem-1515 17 9 node node NOUN fbem-1515 17 10 of of ADP fbem-1515 17 11 the the DET fbem-1515 17 12 base base NOUN fbem-1515 17 13 decision decision NOUN fbem-1515 17 14 tree tree NOUN fbem-1515 17 15 , , PUNCT fbem-1515 17 16 a a DET fbem-1515 17 17 subset subset NOUN fbem-1515 17 18 containing contain VERB fbem-1515 17 19 k k PROPN fbem-1515 17 20 attributes attribute NOUN fbem-1515 17 21 is be AUX fbem-1515 17 22 first first ADV fbem-1515 17 23 randomly randomly ADV fbem-1515 17 24 selected select VERB fbem-1515 17 25 from from ADP fbem-1515 17 26 the the DET fbem-1515 17 27 set set NOUN fbem-1515 17 28 of of ADP fbem-1515 17 29 attributes attribute NOUN fbem-1515 17 30 of of ADP fbem-1515 17 31 that that DET fbem-1515 17 32 node node NOUN fbem-1515 17 33 , , PUNCT fbem-1515 17 34 and and CCONJ fbem-1515 17 35 then then ADV fbem-1515 17 36 an an DET fbem-1515 17 37 optimal optimal ADJ fbem-1515 17 38 attribute attribute NOUN fbem-1515 17 39 is be AUX fbem-1515 17 40 selected select VERB fbem-1515 17 41 from from ADP fbem-1515 17 42 this this DET fbem-1515 17 43 subset subset NOUN fbem-1515 17 44 for for ADP fbem-1515 17 45 division division NOUN fbem-1515 17 46 , , PUNCT fbem-1515 17 47 where where SCONJ fbem-1515 17 48 the the DET fbem-1515 17 49 parameter parameter NOUN fbem-1515 17 50 k k PROPN fbem-1515 17 51 controls control VERB fbem-1515 17 52 the the DET fbem-1515 17 53 degree degree NOUN fbem-1515 17 54 of of ADP fbem-1515 17 55 randomness randomness NOUN fbem-1515 17 56 , , PUNCT fbem-1515 17 57 generally generally ADV fbem-1515 17 58 π‘˜=log2d π‘˜=log2d PUNCT fbem-1515 18 1 [ [ X fbem-1515 18 2 7 7 NUM fbem-1515 18 3 ] ] PUNCT fbem-1515 18 4 . . PUNCT fbem-1515 19 1 in in ADP fbem-1515 19 2 a a DET fbem-1515 19 3 random random ADJ fbem-1515 19 4 forest forest NOUN fbem-1515 19 5 , , PUNCT fbem-1515 19 6 each each DET fbem-1515 19 7 decision decision NOUN fbem-1515 19 8 tree tree NOUN fbem-1515 19 9 is be AUX fbem-1515 19 10 disjoint disjoint ADJ fbem-1515 19 11 , , PUNCT fbem-1515 19 12 and and CCONJ fbem-1515 19 13 the the DET fbem-1515 19 14 final final ADJ fbem-1515 19 15 classification classification NOUN fbem-1515 19 16 result result NOUN fbem-1515 19 17 is be AUX fbem-1515 19 18 determined determine VERB fbem-1515 19 19 by by ADP fbem-1515 19 20 the the DET fbem-1515 19 21 plurality plurality NOUN fbem-1515 19 22 of of ADP fbem-1515 19 23 the the DET fbem-1515 19 24 results result NOUN fbem-1515 19 25 obtained obtain VERB fbem-1515 19 26 from from ADP fbem-1515 19 27 each each DET fbem-1515 19 28 decision decision NOUN fbem-1515 19 29 tree tree NOUN fbem-1515 19 30 . . PUNCT fbem-1515 20 1 2 2 X fbem-1515 20 2 . . X fbem-1515 20 3 introduction introduction NOUN fbem-1515 20 4 : : PUNCT fbem-1515 20 5 business business NOUN fbem-1515 20 6 goal goal NOUN fbem-1515 20 7 loan loan NOUN fbem-1515 20 8 is be AUX fbem-1515 20 9 an an DET fbem-1515 20 10 essential essential ADJ fbem-1515 20 11 business business NOUN fbem-1515 20 12 for for ADP fbem-1515 20 13 many many ADJ fbem-1515 20 14 financial financial ADJ fbem-1515 20 15 institutions institution NOUN fbem-1515 20 16 . . PUNCT fbem-1515 21 1 revenue revenue NOUN fbem-1515 21 2 can can AUX fbem-1515 21 3 be be AUX fbem-1515 21 4 earned earn VERB fbem-1515 21 5 from from ADP fbem-1515 21 6 the the DET fbem-1515 21 7 commission commission NOUN fbem-1515 21 8 fee fee NOUN fbem-1515 21 9 ( ( PUNCT fbem-1515 21 10 interests interest NOUN fbem-1515 21 11 ) ) PUNCT fbem-1515 21 12 on on ADP fbem-1515 21 13 any any DET fbem-1515 21 14 outstanding outstanding ADJ fbem-1515 21 15 loan loan NOUN fbem-1515 21 16 while while SCONJ fbem-1515 21 17 a a DET fbem-1515 21 18 loss loss NOUN fbem-1515 21 19 will will AUX fbem-1515 21 20 be be AUX fbem-1515 21 21 incurred incur VERB fbem-1515 21 22 if if SCONJ fbem-1515 21 23 a a DET fbem-1515 21 24 loan loan NOUN fbem-1515 21 25 defaults default NOUN fbem-1515 21 26 . . PUNCT fbem-1515 22 1 thus thus ADV fbem-1515 22 2 , , PUNCT fbem-1515 22 3 whether whether SCONJ fbem-1515 22 4 to to PART fbem-1515 22 5 approve approve VERB fbem-1515 22 6 a a DET fbem-1515 22 7 loan loan NOUN fbem-1515 22 8 of of ADP fbem-1515 22 9 a a DET fbem-1515 22 10 certain certain ADJ fbem-1515 22 11 amount amount NOUN fbem-1515 22 12 to to ADP fbem-1515 22 13 a a DET fbem-1515 22 14 certain certain ADJ fbem-1515 22 15 customer customer NOUN fbem-1515 22 16 becomes become VERB fbem-1515 22 17 important important ADJ fbem-1515 22 18 for for SCONJ fbem-1515 22 19 such such ADJ fbem-1515 22 20 financial financial ADJ fbem-1515 22 21 institutions institution NOUN fbem-1515 22 22 to to PART fbem-1515 22 23 maximize maximize VERB fbem-1515 22 24 profit profit NOUN fbem-1515 22 25 . . PUNCT fbem-1515 23 1 we we PRON fbem-1515 23 2 hope hope VERB fbem-1515 23 3 to to PART fbem-1515 23 4 build build VERB fbem-1515 23 5 a a DET fbem-1515 23 6 machine machine NOUN fbem-1515 23 7 learning learn VERB fbem-1515 23 8 model model NOUN fbem-1515 23 9 to to PART fbem-1515 23 10 predict predict VERB fbem-1515 23 11 whether whether SCONJ fbem-1515 23 12 a a DET fbem-1515 23 13 customer customer NOUN fbem-1515 23 14 will will AUX fbem-1515 23 15 default default VERB fbem-1515 23 16 in in ADP fbem-1515 23 17 the the DET fbem-1515 23 18 future future NOUN fbem-1515 23 19 , , PUNCT fbem-1515 23 20 given give VERB fbem-1515 23 21 some some DET fbem-1515 23 22 key key ADJ fbem-1515 23 23 information information NOUN fbem-1515 23 24 provided provide VERB fbem-1515 23 25 . . PUNCT fbem-1515 24 1 so so ADV fbem-1515 24 2 that that SCONJ fbem-1515 24 3 the the DET fbem-1515 24 4 institution institution NOUN fbem-1515 24 5 can can AUX fbem-1515 24 6 decide decide VERB fbem-1515 24 7 whether whether SCONJ fbem-1515 24 8 to to PART fbem-1515 24 9 approve approve VERB fbem-1515 24 10 or or CCONJ fbem-1515 24 11 reject reject VERB fbem-1515 24 12 a a DET fbem-1515 24 13 loan loan NOUN fbem-1515 24 14 application application NOUN fbem-1515 24 15 according accord VERB fbem-1515 24 16 to to ADP fbem-1515 24 17 the the DET fbem-1515 24 18 prediction prediction NOUN fbem-1515 24 19 ( ( PUNCT fbem-1515 24 20 approve approve VERB fbem-1515 24 21 if if SCONJ fbem-1515 24 22 non non ADJ fbem-1515 24 23 - - NOUN fbem-1515 24 24 default default NOUN fbem-1515 24 25 and and CCONJ fbem-1515 24 26 vice vice NOUN fbem-1515 24 27 versa versa ADV fbem-1515 24 28 ) ) PUNCT fbem-1515 24 29 . . PUNCT fbem-1515 25 1 our our PRON fbem-1515 25 2 models model NOUN fbem-1515 25 3 target target VERB fbem-1515 25 4 to to PART fbem-1515 25 5 make make VERB fbem-1515 25 6 predictions prediction NOUN fbem-1515 25 7 that that PRON fbem-1515 25 8 maximize maximize VERB fbem-1515 25 9 profits profit NOUN fbem-1515 25 10 . . PUNCT fbem-1515 26 1 and and CCONJ fbem-1515 26 2 we we PRON fbem-1515 26 3 would would AUX fbem-1515 26 4 also also ADV fbem-1515 26 5 like like VERB fbem-1515 26 6 to to PART fbem-1515 26 7 conduct conduct VERB fbem-1515 26 8 further further ADJ fbem-1515 26 9 analysis analysis NOUN fbem-1515 26 10 on on ADP fbem-1515 26 11 our our PRON fbem-1515 26 12 models model NOUN fbem-1515 26 13 to to PART fbem-1515 26 14 gain gain VERB fbem-1515 26 15 some some DET fbem-1515 26 16 insight insight NOUN fbem-1515 26 17 into into ADP fbem-1515 26 18 feature feature NOUN fbem-1515 26 19 importance importance NOUN fbem-1515 26 20 and and CCONJ fbem-1515 26 21 data datum NOUN fbem-1515 26 22 selection selection NOUN fbem-1515 26 23 . . PUNCT fbem-1515 27 1 based base VERB fbem-1515 27 2 on on ADP fbem-1515 27 3 our our PRON fbem-1515 27 4 findings finding NOUN fbem-1515 27 5 , , PUNCT fbem-1515 27 6 we we PRON fbem-1515 27 7 attempt attempt VERB fbem-1515 27 8 to to PART fbem-1515 27 9 propose propose VERB fbem-1515 27 10 some some DET fbem-1515 27 11 explanations explanation NOUN fbem-1515 27 12 and and CCONJ fbem-1515 27 13 business business NOUN fbem-1515 27 14 suggestions suggestion NOUN fbem-1515 27 15 for for ADP fbem-1515 27 16 the the DET fbem-1515 27 17 financial financial ADJ fbem-1515 27 18 institution institution NOUN fbem-1515 27 19 . . PUNCT fbem-1515 28 1 3 3 X fbem-1515 28 2 . . X fbem-1515 28 3 data datum NOUN fbem-1515 28 4 understanding understanding NOUN fbem-1515 28 5 and and CCONJ fbem-1515 28 6 processing processing NOUN fbem-1515 28 7 3.1 3.1 NUM fbem-1515 28 8 . . PUNCT fbem-1515 29 1 dataset dataset VERB fbem-1515 29 2 description description NOUN fbem-1515 29 3 source source NOUN fbem-1515 29 4 : : PUNCT fbem-1515 29 5 https://www.kaggle.com/saurabhbagchi/dishnetwork-hackathon?select=train_dataset.csvw https://www.kaggle.com/saurabhbagchi/dishnetwork-hackathon?select=train_dataset.csvw NOUN fbem-1515 29 6 label label NOUN fbem-1515 29 7 : : PUNCT fbem-1515 29 8 0 0 NUM fbem-1515 29 9 ( ( PUNCT fbem-1515 29 10 not not PART fbem-1515 29 11 default default NOUN fbem-1515 29 12 ) ) PUNCT fbem-1515 29 13 , , PUNCT fbem-1515 29 14 1 1 NUM fbem-1515 29 15 ( ( PUNCT fbem-1515 29 16 default default NOUN fbem-1515 29 17 ) ) PUNCT fbem-1515 29 18 , , PUNCT fbem-1515 29 19 the the DET fbem-1515 29 20 distribution distribution NOUN fbem-1515 29 21 is be AUX fbem-1515 29 22 shown show VERB fbem-1515 29 23 in in ADP fbem-1515 29 24 the the DET fbem-1515 29 25 pie pie NOUN fbem-1515 29 26 chart chart NOUN fbem-1515 29 27 ( ( PUNCT fbem-1515 29 28 exhibit exhibit NOUN fbem-1515 29 29 1 1 NUM fbem-1515 29 30 ) ) PUNCT fbem-1515 29 31 features feature VERB fbem-1515 29 32 : : PUNCT fbem-1515 29 33 there there PRON fbem-1515 29 34 are be VERB fbem-1515 29 35 38 38 NUM fbem-1515 29 36 features feature NOUN fbem-1515 29 37 ( ( PUNCT fbem-1515 29 38 meaning mean VERB fbem-1515 29 39 shown show VERB fbem-1515 29 40 in in ADP fbem-1515 29 41 the the DET fbem-1515 29 42 data data PROPN fbem-1515 29 43 dictionary dictionary ADJ fbem-1515 29 44 / / SYM fbem-1515 29 45 progress progress NOUN fbem-1515 29 46 report report NOUN fbem-1515 29 47 ) ) PUNCT fbem-1515 29 48 with with ADP fbem-1515 29 49 different different ADJ fbem-1515 29 50 missing% missing% PUNCT fbem-1515 29 51 as as SCONJ fbem-1515 29 52 listed list VERB fbem-1515 29 53 below below ADV fbem-1515 29 54 ( ( PUNCT fbem-1515 29 55 exhibit exhibit NOUN fbem-1515 29 56 2 2 NUM fbem-1515 29 57 ) ) PUNCT fbem-1515 29 58 . . PUNCT fbem-1515 30 1 considering consider VERB fbem-1515 30 2 the the DET fbem-1515 30 3 large large ADJ fbem-1515 30 4 missing missing ADJ fbem-1515 30 5 proportion proportion NOUN fbem-1515 30 6 and and CCONJ fbem-1515 30 7 feature feature NOUN fbem-1515 30 8 contribution contribution NOUN fbem-1515 30 9 , , PUNCT fbem-1515 30 10 we we PRON fbem-1515 30 11 have have AUX fbem-1515 30 12 dropped drop VERB fbem-1515 30 13 some some DET fbem-1515 30 14 variables variable NOUN fbem-1515 30 15 at at ADP fbem-1515 30 16 the the DET fbem-1515 30 17 beginning beginning NOUN fbem-1515 30 18 such such ADJ fbem-1515 30 19 as as ADP fbem-1515 30 20 β€˜ ' PUNCT fbem-1515 30 21 own_house_age own_house_age NOUN fbem-1515 30 22 ’ ' PUNCT fbem-1515 30 23 and and CCONJ fbem-1515 30 24 β€˜ ' PUNCT fbem-1515 30 25 social_circle_default social_circle_default X fbem-1515 30 26 ’ ' PUNCT fbem-1515 30 27 . . PUNCT fbem-1515 31 1 as as SCONJ fbem-1515 31 2 β€˜ ' PUNCT fbem-1515 31 3 score_scource_1 score_scource_1 NOUN fbem-1515 31 4 ’ ' PUNCT fbem-1515 31 5 is be AUX fbem-1515 31 6 relatively relatively ADV fbem-1515 31 7 highly highly ADV fbem-1515 31 8 correlated correlate VERB fbem-1515 31 9 to to ADP fbem-1515 31 10 the the DET fbem-1515 31 11 label label NOUN fbem-1515 31 12 , , PUNCT fbem-1515 31 13 we we PRON fbem-1515 31 14 decide decide VERB fbem-1515 31 15 to to PART fbem-1515 31 16 keep keep VERB fbem-1515 31 17 it it PRON fbem-1515 31 18 for for ADP fbem-1515 31 19 now now ADV fbem-1515 31 20 . . PUNCT fbem-1515 32 1 figure figure VERB fbem-1515 32 2 1 1 NUM fbem-1515 32 3 . . PUNCT fbem-1515 33 1 the the DET fbem-1515 33 2 distribution distribution NOUN fbem-1515 33 3 of of ADP fbem-1515 33 4 default default NOUN fbem-1515 33 5 and and CCONJ fbem-1515 33 6 non non ADJ fbem-1515 33 7 - - ADJ fbem-1515 33 8 default default ADJ fbem-1515 33 9 143 143 NUM fbem-1515 33 10 figure figure NOUN fbem-1515 33 11 2 2 NUM fbem-1515 33 12 . . PUNCT fbem-1515 34 1 the the DET fbem-1515 34 2 missing missing ADJ fbem-1515 34 3 percentage percentage NOUN fbem-1515 34 4 of of ADP fbem-1515 34 5 38 38 NUM fbem-1515 34 6 features feature NOUN fbem-1515 34 7 3.2 3.2 NUM fbem-1515 34 8 . . PUNCT fbem-1515 35 1 process process NOUN fbem-1515 35 2 methodology methodology NOUN fbem-1515 35 3 filling fill VERB fbem-1515 35 4 in in ADP fbem-1515 35 5 the the DET fbem-1515 35 6 missing miss VERB fbem-1515 35 7 values value NOUN fbem-1515 35 8 : : PUNCT fbem-1515 35 9 most most ADJ fbem-1515 35 10 of of ADP fbem-1515 35 11 the the DET fbem-1515 35 12 numerical numerical ADJ fbem-1515 35 13 missing miss VERB fbem-1515 35 14 values value NOUN fbem-1515 35 15 are be AUX fbem-1515 35 16 filled fill VERB fbem-1515 35 17 with with ADP fbem-1515 35 18 average average ADJ fbem-1515 35 19 group group NOUN fbem-1515 35 20 by by ADP fbem-1515 35 21 correlated correlate VERB fbem-1515 35 22 features feature NOUN fbem-1515 35 23 most most ADJ fbem-1515 35 24 of of ADP fbem-1515 35 25 the the DET fbem-1515 35 26 categorical categorical ADJ fbem-1515 35 27 missing miss VERB fbem-1515 35 28 values value NOUN fbem-1515 35 29 are be AUX fbem-1515 35 30 filled fill VERB fbem-1515 35 31 with with ADP fbem-1515 35 32 mode mode NOUN fbem-1515 35 33 group group NOUN fbem-1515 35 34 by by ADP fbem-1515 35 35 correlated correlate VERB fbem-1515 35 36 features feature NOUN fbem-1515 35 37 normalization normalization NOUN fbem-1515 35 38 : : PUNCT fbem-1515 35 39 observing observe VERB fbem-1515 35 40 the the DET fbem-1515 35 41 distribution distribution NOUN fbem-1515 35 42 of of ADP fbem-1515 35 43 different different ADJ fbem-1515 35 44 features feature NOUN fbem-1515 35 45 , , PUNCT fbem-1515 35 46 for for ADP fbem-1515 35 47 more more ADV fbem-1515 35 48 centralized centralized ADJ fbem-1515 35 49 and and CCONJ fbem-1515 35 50 normally normally ADV fbem-1515 35 51 distributed distribute VERB fbem-1515 35 52 variables variable NOUN fbem-1515 35 53 , , PUNCT fbem-1515 35 54 we we PRON fbem-1515 35 55 apply apply VERB fbem-1515 35 56 z z NOUN fbem-1515 35 57 - - PUNCT fbem-1515 35 58 normalization normalization NOUN fbem-1515 35 59 ; ; PUNCT fbem-1515 35 60 for for ADP fbem-1515 35 61 those those PRON fbem-1515 35 62 concentrated concentrate VERB fbem-1515 35 63 in in ADP fbem-1515 35 64 a a DET fbem-1515 35 65 small small ADJ fbem-1515 35 66 range range NOUN fbem-1515 35 67 , , PUNCT fbem-1515 35 68 we we PRON fbem-1515 35 69 adopt adopt VERB fbem-1515 35 70 log log NOUN fbem-1515 35 71 transformation transformation NOUN fbem-1515 35 72 . . PUNCT fbem-1515 36 1 ( ( PUNCT fbem-1515 36 2 exhibit exhibit VERB fbem-1515 36 3 3 3 NUM fbem-1515 36 4 ) ) PUNCT fbem-1515 36 5 figure figure NOUN fbem-1515 36 6 3 3 NUM fbem-1515 36 7 . . PUNCT fbem-1515 37 1 the the DET fbem-1515 37 2 visualization visualization NOUN fbem-1515 37 3 of of ADP fbem-1515 37 4 concentrated concentrated ADJ fbem-1515 37 5 variables variable NOUN fbem-1515 37 6 figure figure VERB fbem-1515 37 7 4 4 NUM fbem-1515 37 8 . . PUNCT fbem-1515 38 1 the the DET fbem-1515 38 2 distribution distribution NOUN fbem-1515 38 3 of of ADP fbem-1515 38 4 default default NOUN fbem-1515 38 5 and and CCONJ fbem-1515 38 6 non non ADJ fbem-1515 38 7 - - NOUN fbem-1515 38 8 default default NOUN fbem-1515 38 9 after after ADP fbem-1515 38 10 dropping drop VERB fbem-1515 38 11 some some DET fbem-1515 38 12 non non ADJ fbem-1515 38 13 - - NOUN fbem-1515 38 14 default default ADJ fbem-1515 38 15 records record NOUN fbem-1515 38 16 with with ADP fbem-1515 38 17 β€œ " PUNCT fbem-1515 38 18 na na AUX fbem-1515 38 19 ” " PUNCT fbem-1515 38 20 constricting constrict VERB fbem-1515 38 21 a a DET fbem-1515 38 22 new new ADJ fbem-1515 38 23 data datum NOUN fbem-1515 38 24 set set NOUN fbem-1515 38 25 : : PUNCT fbem-1515 38 26 we we PRON fbem-1515 38 27 have have AUX fbem-1515 38 28 found find VERB fbem-1515 38 29 that that SCONJ fbem-1515 38 30 β€˜ ' PUNCT fbem-1515 38 31 default default NOUN fbem-1515 38 32 ’ ' PUNCT fbem-1515 38 33 records record NOUN fbem-1515 38 34 take take VERB fbem-1515 38 35 around around ADV fbem-1515 38 36 8.08 8.08 NUM fbem-1515 38 37 % % NOUN fbem-1515 38 38 in in ADP fbem-1515 38 39 the the DET fbem-1515 38 40 raw raw ADJ fbem-1515 38 41 data datum NOUN fbem-1515 38 42 . . PUNCT fbem-1515 39 1 ( ( PUNCT fbem-1515 39 2 exhibit exhibit VERB fbem-1515 39 3 4 4 NUM fbem-1515 39 4 ) ) PUNCT fbem-1515 39 5 to to PART fbem-1515 39 6 help help VERB fbem-1515 39 7 model model VERB fbem-1515 39 8 better well ADV fbem-1515 39 9 recognize recognize VERB fbem-1515 39 10 the the DET fbem-1515 39 11 default default NOUN fbem-1515 39 12 patterns pattern NOUN fbem-1515 39 13 , , PUNCT fbem-1515 39 14 we we PRON fbem-1515 39 15 try try VERB fbem-1515 39 16 to to PART fbem-1515 39 17 drop drop VERB fbem-1515 39 18 the the DET fbem-1515 39 19 non non ADJ fbem-1515 39 20 - - NOUN fbem-1515 39 21 default default ADJ fbem-1515 39 22 records record NOUN fbem-1515 39 23 with with ADP fbem-1515 39 24 β€˜ ' PUNCT fbem-1515 39 25 na na NOUN fbem-1515 39 26 ’ ' PUNCT fbem-1515 39 27 features feature NOUN fbem-1515 39 28 and and CCONJ fbem-1515 39 29 get get VERB fbem-1515 39 30 a a DET fbem-1515 39 31 dataset dataset NOUN fbem-1515 39 32 with with ADP fbem-1515 39 33 80.37 80.37 NUM fbem-1515 39 34 % % NOUN fbem-1515 39 35 defaults default NOUN fbem-1515 39 36 ( ( PUNCT fbem-1515 39 37 total total ADJ fbem-1515 39 38 12249 12249 NUM fbem-1515 39 39 records record NOUN fbem-1515 39 40 ) ) PUNCT fbem-1515 39 41 . . PUNCT fbem-1515 40 1 [ [ X fbem-1515 40 2 please please INTJ fbem-1515 40 3 refer refer VERB fbem-1515 40 4 to to ADP fbem-1515 40 5 data datum NOUN fbem-1515 40 6 _ _ PRON fbem-1515 40 7 processing_version1 processing_version1 PROPN fbem-1515 40 8 and and CCONJ fbem-1515 40 9 data datum NOUN fbem-1515 40 10 processing_version2(d processing_version2(d PROPN fbem-1515 40 11 ) ) PUNCT fbem-1515 40 12 ] ] PUNCT fbem-1515 41 1 144 144 NUM fbem-1515 42 1 and and CCONJ fbem-1515 42 2 we we PRON fbem-1515 42 3 build build VERB fbem-1515 42 4 each each DET fbem-1515 42 5 type type NOUN fbem-1515 42 6 of of ADP fbem-1515 42 7 model model NOUN fbem-1515 42 8 on on ADP fbem-1515 42 9 both both DET fbem-1515 42 10 datasets dataset NOUN fbem-1515 42 11 : : PUNCT fbem-1515 42 12 ( ( PUNCT fbem-1515 42 13 1)the 1)the DET fbem-1515 42 14 original original ADJ fbem-1515 42 15 one one NUM fbem-1515 42 16 with with ADP fbem-1515 42 17 all all DET fbem-1515 42 18 features feature NOUN fbem-1515 42 19 : : PUNCT fbem-1515 42 20 train train NOUN fbem-1515 42 21 ( ( PUNCT fbem-1515 42 22 2 2 X fbem-1515 42 23 ) ) PUNCT fbem-1515 42 24 the the DET fbem-1515 42 25 one one NOUN fbem-1515 42 26 with with ADP fbem-1515 42 27 non non NOUN fbem-1515 42 28 - - NOUN fbem-1515 42 29 default default NOUN fbem-1515 42 30 na na PART fbem-1515 42 31 dropped drop VERB fbem-1515 42 32 : : PUNCT fbem-1515 42 33 train_1 train_1 PROPN fbem-1515 42 34 4 4 NUM fbem-1515 42 35 . . PUNCT fbem-1515 42 36 model model NOUN fbem-1515 42 37 training training NOUN fbem-1515 42 38 and and CCONJ fbem-1515 42 39 evaluation evaluation NOUN fbem-1515 42 40 4.1 4.1 NUM fbem-1515 42 41 . . PUNCT fbem-1515 43 1 cost cost NOUN fbem-1515 43 2 and and CCONJ fbem-1515 43 3 benefit benefit VERB fbem-1515 43 4 analysis analysis NOUN fbem-1515 43 5 assumption assumption NOUN fbem-1515 43 6 commission commission NOUN fbem-1515 43 7 fee fee NOUN fbem-1515 43 8 : : PUNCT fbem-1515 43 9 6 6 NUM fbem-1515 43 10 % % NOUN fbem-1515 43 11 ( ( PUNCT fbem-1515 43 12 for for ADP fbem-1515 43 13 non non ADJ fbem-1515 43 14 - - ADJ fbem-1515 43 15 default default ADJ fbem-1515 43 16 customers customer NOUN fbem-1515 43 17 ) ) PUNCT fbem-1515 43 18 exposure exposure NOUN fbem-1515 43 19 at at ADP fbem-1515 43 20 default default NOUN fbem-1515 43 21 ( ( PUNCT fbem-1515 43 22 ead ead ADJ fbem-1515 43 23 ) ) PUNCT fbem-1515 43 24 or or CCONJ fbem-1515 43 25 loan loan NOUN fbem-1515 43 26 amount amount NOUN fbem-1515 43 27 per per ADP fbem-1515 43 28 account account NOUN fbem-1515 43 29 : : PUNCT fbem-1515 43 30 median median NOUN fbem-1515 43 31 of of ADP fbem-1515 43 32 the the DET fbem-1515 43 33 credit credit NOUN fbem-1515 43 34 amount amount NOUN fbem-1515 43 35 loss loss NOUN fbem-1515 43 36 given give VERB fbem-1515 43 37 default default NOUN fbem-1515 43 38 ( ( PUNCT fbem-1515 43 39 lgd lgd PROPN fbem-1515 43 40 ): ): PUNCT fbem-1515 43 41 55 55 NUM fbem-1515 43 42 % % NOUN fbem-1515 43 43 ( ( PUNCT fbem-1515 43 44 refer refer VERB fbem-1515 43 45 to to ADP fbem-1515 43 46 the the DET fbem-1515 43 47 hkma hkma NOUN fbem-1515 43 48 45 45 NUM fbem-1515 43 49 % % NOUN fbem-1515 43 50 recovery recovery NOUN fbem-1515 43 51 for for ADP fbem-1515 43 52 unsecured unsecured ADJ fbem-1515 43 53 loan loan NOUN fbem-1515 43 54 ) ) PUNCT fbem-1515 43 55 financial financial ADJ fbem-1515 43 56 institution institution NOUN fbem-1515 43 57 will will AUX fbem-1515 43 58 reject reject VERB fbem-1515 43 59 all all DET fbem-1515 43 60 customers customer NOUN fbem-1515 43 61 with with ADP fbem-1515 43 62 β€˜ ' PUNCT fbem-1515 43 63 default default NOUN fbem-1515 43 64 ’ ' PUNCT fbem-1515 43 65 prediction prediction NOUN fbem-1515 43 66 opportunity opportunity NOUN fbem-1515 43 67 cost cost VERB fbem-1515 43 68 cost= cost= NOUN fbem-1515 43 69 # # NOUN fbem-1515 43 70 ( ( PUNCT fbem-1515 43 71 false false ADJ fbem-1515 43 72 positive positive ADJ fbem-1515 43 73 ) ) PUNCT fbem-1515 43 74 * * PUNCT fbem-1515 43 75 commission commission NOUN fbem-1515 43 76 fee fee NOUN fbem-1515 43 77 * * PUNCT fbem-1515 43 78 loan loan NOUN fbem-1515 43 79 amount amount NOUN fbem-1515 43 80 + + CCONJ fbem-1515 43 81 # # NOUN fbem-1515 43 82 ( ( PUNCT fbem-1515 43 83 false false ADJ fbem-1515 43 84 negative negative NOUN fbem-1515 43 85 ) ) PUNCT fbem-1515 43 86 * * PUNCT fbem-1515 43 87 ead*lgd ead*lgd ADJ fbem-1515 43 88 goal goal NOUN fbem-1515 43 89 : : PUNCT fbem-1515 43 90 as as SCONJ fbem-1515 43 91 the the DET fbem-1515 43 92 financial financial ADJ fbem-1515 43 93 institution institution NOUN fbem-1515 43 94 would would AUX fbem-1515 43 95 like like VERB fbem-1515 43 96 to to PART fbem-1515 43 97 maximize maximize VERB fbem-1515 43 98 their their PRON fbem-1515 43 99 profits profit NOUN fbem-1515 43 100 from from ADP fbem-1515 43 101 the the DET fbem-1515 43 102 car car NOUN fbem-1515 43 103 loans loan NOUN fbem-1515 43 104 , , PUNCT fbem-1515 43 105 we we PRON fbem-1515 43 106 will will AUX fbem-1515 43 107 select select VERB fbem-1515 43 108 the the DET fbem-1515 43 109 optimal optimal ADJ fbem-1515 43 110 model model NOUN fbem-1515 43 111 threshold threshold NOUN fbem-1515 43 112 to to PART fbem-1515 43 113 minimize minimize VERB fbem-1515 43 114 the the DET fbem-1515 43 115 opportunity opportunity NOUN fbem-1515 43 116 cost cost NOUN fbem-1515 43 117 in in ADP fbem-1515 43 118 the the DET fbem-1515 43 119 later later ADJ fbem-1515 43 120 stage stage NOUN fbem-1515 43 121 . . PUNCT fbem-1515 44 1 4.2 4.2 NUM fbem-1515 44 2 . . PUNCT fbem-1515 45 1 decision decision NOUN fbem-1515 45 2 tree tree NOUN fbem-1515 45 3 data datum NOUN fbem-1515 45 4 source source NOUN fbem-1515 45 5 : : PUNCT fbem-1515 45 6 train_cleaned.csv(β€œtrain train_cleaned.csv(β€œtrain NOUN fbem-1515 45 7 ” " PUNCT fbem-1515 45 8 ) ) PUNCT fbem-1515 45 9 , , PUNCT fbem-1515 45 10 train_1_cleaned.csv train_1_cleaned.csv NUM fbem-1515 45 11 ( ( PUNCT fbem-1515 45 12 β€œ " PUNCT fbem-1515 45 13 train_1 train_1 PROPN fbem-1515 45 14 ” " PUNCT fbem-1515 45 15 ) ) PUNCT fbem-1515 46 1 we we PRON fbem-1515 46 2 prepared prepare VERB fbem-1515 46 3 two two NUM fbem-1515 46 4 datasets dataset NOUN fbem-1515 46 5 and and CCONJ fbem-1515 46 6 decided decide VERB fbem-1515 46 7 to to PART fbem-1515 46 8 build build VERB fbem-1515 46 9 one one NUM fbem-1515 46 10 models model NOUN fbem-1515 46 11 on on ADP fbem-1515 46 12 each each PRON fbem-1515 46 13 , , PUNCT fbem-1515 46 14 then then ADV fbem-1515 46 15 we we PRON fbem-1515 46 16 would would AUX fbem-1515 46 17 compare compare VERB fbem-1515 46 18 the the DET fbem-1515 46 19 two two NUM fbem-1515 46 20 models model NOUN fbem-1515 46 21 and and CCONJ fbem-1515 46 22 adopt adopt VERB fbem-1515 46 23 the the DET fbem-1515 46 24 better well ADV fbem-1515 46 25 - - PUNCT fbem-1515 46 26 performing perform VERB fbem-1515 46 27 one one NUM fbem-1515 46 28 . . PUNCT fbem-1515 47 1 the the DET fbem-1515 47 2 β€˜ ' PUNCT fbem-1515 47 3 train train NOUN fbem-1515 47 4 ’ ' PUNCT fbem-1515 47 5 dataset dataset NOUN fbem-1515 47 6 is be AUX fbem-1515 47 7 the the DET fbem-1515 47 8 original original ADJ fbem-1515 47 9 one one NOUN fbem-1515 47 10 which which PRON fbem-1515 47 11 contains contain VERB fbem-1515 47 12 all all DET fbem-1515 47 13 the the DET fbem-1515 47 14 features feature NOUN fbem-1515 47 15 , , PUNCT fbem-1515 47 16 while while SCONJ fbem-1515 47 17 the the DET fbem-1515 47 18 β€˜ ' PUNCT fbem-1515 47 19 train_1 train_1 NOUN fbem-1515 47 20 ’ ' PUNCT fbem-1515 47 21 dataset dataset NOUN fbem-1515 47 22 is be AUX fbem-1515 47 23 the the DET fbem-1515 47 24 one one NUM fbem-1515 47 25 whose whose DET fbem-1515 47 26 non non ADJ fbem-1515 47 27 - - ADJ fbem-1515 47 28 default default ADJ fbem-1515 47 29 examples example NOUN fbem-1515 47 30 with with ADP fbem-1515 47 31 β€œ " PUNCT fbem-1515 47 32 na na NOUN fbem-1515 47 33 ” " PUNCT fbem-1515 47 34 values value NOUN fbem-1515 47 35 have have AUX fbem-1515 47 36 been be AUX fbem-1515 47 37 dropped drop VERB fbem-1515 47 38 . . PUNCT fbem-1515 48 1 specific specific ADJ fbem-1515 48 2 data datum NOUN fbem-1515 48 3 processing processing NOUN fbem-1515 48 4 : : PUNCT fbem-1515 48 5 discretization discretization NOUN fbem-1515 48 6 : : PUNCT fbem-1515 48 7 as as SCONJ fbem-1515 48 8 decision decision NOUN fbem-1515 48 9 tree tree NOUN fbem-1515 48 10 models model NOUN fbem-1515 48 11 tend tend VERB fbem-1515 48 12 to to PART fbem-1515 48 13 behave behave VERB fbem-1515 48 14 better well ADV fbem-1515 48 15 on on ADP fbem-1515 48 16 categorical categorical ADJ fbem-1515 48 17 data datum NOUN fbem-1515 48 18 , , PUNCT fbem-1515 48 19 we we PRON fbem-1515 48 20 examined examine VERB fbem-1515 48 21 some some DET fbem-1515 48 22 numerical numerical ADJ fbem-1515 48 23 data datum NOUN fbem-1515 48 24 . . PUNCT fbem-1515 49 1 and and CCONJ fbem-1515 49 2 for for ADP fbem-1515 49 3 two two NUM fbem-1515 49 4 features feature NOUN fbem-1515 49 5 : : PUNCT fbem-1515 49 6 age_days age_day NOUN fbem-1515 49 7 and and CCONJ fbem-1515 49 8 credit_amount credit_amount PROPN fbem-1515 49 9 , , PUNCT fbem-1515 49 10 we we PRON fbem-1515 49 11 conducted conduct VERB fbem-1515 49 12 discretization discretization NOUN fbem-1515 49 13 and and CCONJ fbem-1515 49 14 converted convert VERB fbem-1515 49 15 them they PRON fbem-1515 49 16 to to ADP fbem-1515 49 17 dummies dummy NOUN fbem-1515 49 18 . . PUNCT fbem-1515 50 1 intuitively intuitively ADV fbem-1515 50 2 , , PUNCT fbem-1515 50 3 they they PRON fbem-1515 50 4 are be AUX fbem-1515 50 5 probably probably ADV fbem-1515 50 6 very very ADV fbem-1515 50 7 important important ADJ fbem-1515 50 8 in in ADP fbem-1515 50 9 predicting predict VERB fbem-1515 50 10 the the DET fbem-1515 50 11 results result NOUN fbem-1515 50 12 and and CCONJ fbem-1515 50 13 are be AUX fbem-1515 50 14 common common ADJ fbem-1515 50 15 targets target NOUN fbem-1515 50 16 for for ADP fbem-1515 50 17 discretization discretization NOUN fbem-1515 50 18 in in ADP fbem-1515 50 19 practice practice NOUN fbem-1515 50 20 . . PUNCT fbem-1515 51 1 while while SCONJ fbem-1515 51 2 statistically statistically ADV fbem-1515 51 3 , , PUNCT fbem-1515 51 4 those those DET fbem-1515 51 5 2 2 NUM fbem-1515 51 6 features feature NOUN fbem-1515 51 7 have have VERB fbem-1515 51 8 long long ADJ fbem-1515 51 9 - - PUNCT fbem-1515 51 10 tail tail NOUN fbem-1515 51 11 issues issue NOUN fbem-1515 51 12 , , PUNCT fbem-1515 51 13 so so SCONJ fbem-1515 51 14 that that SCONJ fbem-1515 51 15 we we PRON fbem-1515 51 16 would would AUX fbem-1515 51 17 like like VERB fbem-1515 51 18 to to ADP fbem-1515 51 19 further further ADJ fbem-1515 51 20 process process NOUN fbem-1515 51 21 . . PUNCT fbem-1515 52 1 hyper hyper ADJ fbem-1515 52 2 - - ADJ fbem-1515 52 3 tunning tunning NOUN fbem-1515 52 4 : : PUNCT fbem-1515 52 5 our our PRON fbem-1515 52 6 major major ADJ fbem-1515 52 7 target target NOUN fbem-1515 52 8 for for ADP fbem-1515 52 9 hyper hyper NOUN fbem-1515 52 10 - - ADJ fbem-1515 52 11 tuning tuning NOUN fbem-1515 52 12 of of ADP fbem-1515 52 13 the the DET fbem-1515 52 14 decision decision NOUN fbem-1515 52 15 tree tree NOUN fbem-1515 52 16 is be AUX fbem-1515 52 17 to to PART fbem-1515 52 18 eliminate eliminate VERB fbem-1515 52 19 the the DET fbem-1515 52 20 over over ADV fbem-1515 52 21 - - PUNCT fbem-1515 52 22 fitting fit VERB fbem-1515 52 23 issue issue NOUN fbem-1515 52 24 . . PUNCT fbem-1515 53 1 so so ADV fbem-1515 53 2 , , PUNCT fbem-1515 53 3 we we PRON fbem-1515 53 4 picked pick VERB fbem-1515 53 5 2 2 NUM fbem-1515 53 6 hyperparameters hyperparameter NOUN fbem-1515 53 7 : : PUNCT fbem-1515 53 8 ' ' PUNCT fbem-1515 53 9 max_depth max_depth X fbem-1515 53 10 ' ' PUNCT fbem-1515 53 11 and and CCONJ fbem-1515 53 12 ' ' PUNCT fbem-1515 53 13 min_samples_split min_samples_split ADJ fbem-1515 53 14 ' ' PUNCT fbem-1515 53 15 and and CCONJ fbem-1515 53 16 did do VERB fbem-1515 53 17 a a DET fbem-1515 53 18 grid grid NOUN fbem-1515 53 19 search search NOUN fbem-1515 53 20 for for ADP fbem-1515 53 21 cross cross NOUN fbem-1515 53 22 validation validation NOUN fbem-1515 53 23 with with ADP fbem-1515 53 24 cv=10 cv=10 PROPN fbem-1515 53 25 . . PUNCT fbem-1515 54 1 ( ( PUNCT fbem-1515 54 2 grid grid NOUN fbem-1515 54 3 search search NOUN fbem-1515 54 4 range range NOUN fbem-1515 54 5 : : PUNCT fbem-1515 54 6 max_depth= max_depth= NOUN fbem-1515 54 7 [ [ PUNCT fbem-1515 54 8 1 1 NUM fbem-1515 54 9 , , PUNCT fbem-1515 54 10 2 2 NUM fbem-1515 54 11 , , PUNCT fbem-1515 54 12 3 3 NUM fbem-1515 54 13 , , PUNCT fbem-1515 54 14 4 4 NUM fbem-1515 54 15 , , PUNCT fbem-1515 54 16 5 5 NUM fbem-1515 54 17 , , PUNCT fbem-1515 54 18 6 6 NUM fbem-1515 54 19 , , PUNCT fbem-1515 54 20 7 7 NUM fbem-1515 54 21 , , PUNCT fbem-1515 54 22 8 8 NUM fbem-1515 54 23 , , PUNCT fbem-1515 54 24 9 9 NUM fbem-1515 54 25 , , PUNCT fbem-1515 54 26 10 10 NUM fbem-1515 54 27 ] ] PUNCT fbem-1515 54 28 , , PUNCT fbem-1515 54 29 ' ' PUNCT fbem-1515 54 30 min_samples_split min_samples_split ADJ fbem-1515 54 31 ' ' PUNCT fbem-1515 54 32 : : PUNCT fbem-1515 55 1 [ [ X fbem-1515 55 2 2 2 NUM fbem-1515 55 3 , , PUNCT fbem-1515 55 4 7 7 NUM fbem-1515 55 5 , , PUNCT fbem-1515 55 6 12 12 NUM fbem-1515 55 7 , , PUNCT fbem-1515 55 8 17 17 NUM fbem-1515 55 9 , , PUNCT fbem-1515 55 10 22 22 NUM fbem-1515 55 11 , , PUNCT fbem-1515 55 12 27 27 NUM fbem-1515 55 13 , , PUNCT fbem-1515 55 14 32 32 NUM fbem-1515 55 15 , , PUNCT fbem-1515 55 16 37 37 NUM fbem-1515 55 17 , , PUNCT fbem-1515 55 18 42 42 NUM fbem-1515 55 19 , , PUNCT fbem-1515 55 20 47 47 NUM fbem-1515 55 21 , , PUNCT fbem-1515 55 22 52 52 NUM fbem-1515 55 23 , , PUNCT fbem-1515 55 24 57 57 NUM fbem-1515 55 25 , , PUNCT fbem-1515 55 26 62 62 NUM fbem-1515 55 27 , , PUNCT fbem-1515 55 28 67 67 NUM fbem-1515 55 29 , , PUNCT fbem-1515 55 30 72 72 NUM fbem-1515 55 31 , , PUNCT fbem-1515 55 32 77 77 NUM fbem-1515 55 33 , , PUNCT fbem-1515 55 34 82 82 NUM fbem-1515 55 35 , , PUNCT fbem-1515 55 36 87 87 NUM fbem-1515 55 37 , , PUNCT fbem-1515 55 38 92 92 NUM fbem-1515 55 39 , , PUNCT fbem-1515 55 40 97 97 NUM fbem-1515 55 41 ] ] PUNCT fbem-1515 55 42 ) ) PUNCT fbem-1515 55 43 as as ADP fbem-1515 55 44 the the DET fbem-1515 55 45 major major ADJ fbem-1515 55 46 target target NOUN fbem-1515 55 47 of of ADP fbem-1515 55 48 this this DET fbem-1515 55 49 model model NOUN fbem-1515 55 50 is be AUX fbem-1515 55 51 to to PART fbem-1515 55 52 maximize maximize VERB fbem-1515 55 53 the the DET fbem-1515 55 54 profit profit NOUN fbem-1515 55 55 , , PUNCT fbem-1515 55 56 we we PRON fbem-1515 55 57 care care VERB fbem-1515 55 58 more more ADJ fbem-1515 55 59 about about ADP fbem-1515 55 60 the the DET fbem-1515 55 61 true true ADJ fbem-1515 55 62 positive positive ADJ fbem-1515 55 63 rate rate NOUN fbem-1515 55 64 and and CCONJ fbem-1515 55 65 false false ADJ fbem-1515 55 66 positive positive ADJ fbem-1515 55 67 rate rate NOUN fbem-1515 55 68 , , PUNCT fbem-1515 55 69 and and CCONJ fbem-1515 55 70 auc auc VERB fbem-1515 55 71 more more ADJ fbem-1515 55 72 than than ADP fbem-1515 55 73 sole sole ADJ fbem-1515 55 74 accuracy accuracy NOUN fbem-1515 55 75 , , PUNCT fbem-1515 55 76 when when SCONJ fbem-1515 55 77 doing do VERB fbem-1515 55 78 the the DET fbem-1515 55 79 grid grid NOUN fbem-1515 55 80 search search NOUN fbem-1515 55 81 , , PUNCT fbem-1515 55 82 we we PRON fbem-1515 55 83 set set VERB fbem-1515 55 84 the the DET fbem-1515 55 85 scoring scoring NOUN fbem-1515 55 86 to to PART fbem-1515 55 87 be be AUX fbem-1515 55 88 roc roc NOUN fbem-1515 55 89 - - PUNCT fbem-1515 55 90 auc auc NOUN fbem-1515 55 91 . . PUNCT fbem-1515 56 1 and and CCONJ fbem-1515 56 2 we we PRON fbem-1515 56 3 eventually eventually ADV fbem-1515 56 4 adopted adopt VERB fbem-1515 56 5 the the DET fbem-1515 56 6 best good ADJ fbem-1515 56 7 parameters parameter NOUN fbem-1515 56 8 output output NOUN fbem-1515 56 9 of of ADP fbem-1515 56 10 max_depth=7 max_depth=7 PROPN fbem-1515 56 11 and and CCONJ fbem-1515 56 12 min_sample_split=72 min_sample_split=72 PROPN fbem-1515 56 13 for for ADP fbem-1515 56 14 model model NOUN fbem-1515 56 15 dt_model dt_model NOUN fbem-1515 56 16 ( ( PUNCT fbem-1515 56 17 train train NOUN fbem-1515 56 18 ) ) PUNCT fbem-1515 56 19 max_depth=8 max_depth=8 PROPN fbem-1515 56 20 and and CCONJ fbem-1515 56 21 min_sample_split=72 min_sample_split=72 NOUN fbem-1515 56 22 for for ADP fbem-1515 56 23 model model NOUN fbem-1515 56 24 dt_model_1 dt_model_1 PROPN fbem-1515 56 25 ( ( PUNCT fbem-1515 56 26 train_1 train_1 NOUN fbem-1515 56 27 ) ) PUNCT fbem-1515 56 28 model model NOUN fbem-1515 56 29 training training NOUN fbem-1515 56 30 : : PUNCT fbem-1515 56 31 we we PRON fbem-1515 56 32 have have AUX fbem-1515 56 33 built build VERB fbem-1515 56 34 our our PRON fbem-1515 56 35 model model NOUN fbem-1515 56 36 on on ADP fbem-1515 56 37 two two NUM fbem-1515 56 38 data data NOUN fbem-1515 56 39 sets set NOUN fbem-1515 56 40 : : PUNCT fbem-1515 56 41 train_1_cleaned train_1_cleane VERB fbem-1515 56 42 ( ( PUNCT fbem-1515 56 43 train_1_df train_1_df PROPN fbem-1515 56 44 ) ) PUNCT fbem-1515 56 45 and and CCONJ fbem-1515 56 46 train_cleaned train_cleane VERB fbem-1515 56 47 ( ( PUNCT fbem-1515 56 48 train_df train_df PROPN fbem-1515 56 49 ) ) PUNCT fbem-1515 56 50 , , PUNCT fbem-1515 56 51 and and CCONJ fbem-1515 56 52 would would AUX fbem-1515 56 53 decide decide VERB fbem-1515 56 54 which which DET fbem-1515 56 55 one one NOUN fbem-1515 56 56 to to PART fbem-1515 56 57 use use VERB fbem-1515 56 58 according accord VERB fbem-1515 56 59 to to ADP fbem-1515 56 60 their their PRON fbem-1515 56 61 performance performance NOUN fbem-1515 56 62 on on ADP fbem-1515 56 63 the the DET fbem-1515 56 64 same same ADJ fbem-1515 56 65 test test NOUN fbem-1515 56 66 set set VERB fbem-1515 56 67 . . PUNCT fbem-1515 57 1 we we PRON fbem-1515 57 2 split split VERB fbem-1515 57 3 the the DET fbem-1515 57 4 train_1 train_1 NOUN fbem-1515 57 5 and and CCONJ fbem-1515 57 6 train train VERB fbem-1515 57 7 into into ADP fbem-1515 57 8 train train NOUN fbem-1515 57 9 and and CCONJ fbem-1515 57 10 test test NOUN fbem-1515 57 11 set set VERB fbem-1515 57 12 . . PUNCT fbem-1515 58 1 considering consider VERB fbem-1515 58 2 we we PRON fbem-1515 58 3 would would AUX fbem-1515 58 4 need need VERB fbem-1515 58 5 to to PART fbem-1515 58 6 conduct conduct VERB fbem-1515 58 7 a a DET fbem-1515 58 8 cost cost NOUN fbem-1515 58 9 - - PUNCT fbem-1515 58 10 benefit benefit NOUN fbem-1515 58 11 analysis analysis NOUN fbem-1515 58 12 to to PART fbem-1515 58 13 determine determine VERB fbem-1515 58 14 an an DET fbem-1515 58 15 optimal optimal ADJ fbem-1515 58 16 threshold threshold NOUN fbem-1515 58 17 , , PUNCT fbem-1515 58 18 we we PRON fbem-1515 58 19 further far ADV fbem-1515 58 20 split split VERB fbem-1515 58 21 the the DET fbem-1515 58 22 train train NOUN fbem-1515 58 23 set set VERB fbem-1515 58 24 in in ADP fbem-1515 58 25 train train NOUN fbem-1515 58 26 into into ADP fbem-1515 58 27 a a DET fbem-1515 58 28 sub sub ADJ fbem-1515 58 29 - - ADJ fbem-1515 58 30 train train ADJ fbem-1515 58 31 set set NOUN fbem-1515 58 32 and and CCONJ fbem-1515 58 33 a a DET fbem-1515 58 34 validation validation NOUN fbem-1515 58 35 set set NOUN fbem-1515 58 36 ( ( PUNCT fbem-1515 58 37 exhibit exhibit NOUN fbem-1515 58 38 5 5 NUM fbem-1515 58 39 ) ) PUNCT fbem-1515 58 40 . . PUNCT fbem-1515 59 1 figure figure NOUN fbem-1515 59 2 5 5 NUM fbem-1515 59 3 . . PUNCT fbem-1515 59 4 split split VERB fbem-1515 59 5 the the DET fbem-1515 59 6 train train NOUN fbem-1515 59 7 set set VERB fbem-1515 59 8 in in ADP fbem-1515 59 9 train train NOUN fbem-1515 59 10 into into ADP fbem-1515 59 11 a a DET fbem-1515 59 12 sub sub ADJ fbem-1515 59 13 - - ADJ fbem-1515 59 14 train train ADJ fbem-1515 59 15 set set NOUN fbem-1515 59 16 and and CCONJ fbem-1515 59 17 a a DET fbem-1515 59 18 validation validation NOUN fbem-1515 59 19 set set NOUN fbem-1515 59 20 then then ADV fbem-1515 59 21 we we PRON fbem-1515 59 22 built build VERB fbem-1515 59 23 a a DET fbem-1515 59 24 decision decision NOUN fbem-1515 59 25 tree tree NOUN fbem-1515 59 26 model model NOUN fbem-1515 59 27 dt_model_1 dt_model_1 PROPN fbem-1515 59 28 on on ADP fbem-1515 59 29 train_1_df train_1_df PROPN fbem-1515 59 30 and and CCONJ fbem-1515 59 31 fit fit ADJ fbem-1515 59 32 test test NOUN fbem-1515 59 33 set set NOUN fbem-1515 59 34 of of ADP fbem-1515 59 35 train_df train_df PROPN fbem-1515 59 36 into into ADP fbem-1515 59 37 the the DET fbem-1515 59 38 model model NOUN fbem-1515 59 39 and and CCONJ fbem-1515 59 40 got get VERB fbem-1515 59 41 the the DET fbem-1515 59 42 result result NOUN fbem-1515 59 43 accuracy accuracy NOUN fbem-1515 59 44 of of ADP fbem-1515 59 45 only only ADV fbem-1515 59 46 0.26 0.26 NUM fbem-1515 59 47 figure figure NOUN fbem-1515 59 48 6 6 NUM fbem-1515 59 49 . . PUNCT fbem-1515 60 1 the the DET fbem-1515 60 2 result result NOUN fbem-1515 60 3 table table NOUN fbem-1515 60 4 of of ADP fbem-1515 60 5 decision decision NOUN fbem-1515 60 6 tree tree NOUN fbem-1515 60 7 model model NOUN fbem-1515 60 8 built build VERB fbem-1515 60 9 based base VERB fbem-1515 60 10 on on ADP fbem-1515 60 11 train_1_df train_1_df PROPN fbem-1515 60 12 due due ADP fbem-1515 60 13 to to ADP fbem-1515 60 14 the the DET fbem-1515 60 15 very very ADV fbem-1515 60 16 poor poor ADJ fbem-1515 60 17 performance performance NOUN fbem-1515 60 18 of of ADP fbem-1515 60 19 dt_model_1 dt_model_1 NOUN fbem-1515 60 20 in in ADP fbem-1515 60 21 terms term NOUN fbem-1515 60 22 of of ADP fbem-1515 60 23 accuracy accuracy NOUN fbem-1515 60 24 , , PUNCT fbem-1515 60 25 we we PRON fbem-1515 60 26 decided decide VERB fbem-1515 60 27 to to PART fbem-1515 60 28 quit quit VERB fbem-1515 60 29 the the DET fbem-1515 60 30 model model NOUN fbem-1515 60 31 . . PUNCT fbem-1515 61 1 besides besides SCONJ fbem-1515 61 2 , , PUNCT fbem-1515 61 3 we we PRON fbem-1515 61 4 also also ADV fbem-1515 61 5 noticed notice VERB fbem-1515 61 6 that that SCONJ fbem-1515 61 7 this this DET fbem-1515 61 8 model model NOUN fbem-1515 61 9 has have VERB fbem-1515 61 10 an an DET fbem-1515 61 11 issue issue NOUN fbem-1515 61 12 of of ADP fbem-1515 61 13 inadequate inadequate ADJ fbem-1515 61 14 data datum NOUN fbem-1515 61 15 size size NOUN fbem-1515 61 16 ( ( PUNCT fbem-1515 61 17 number number NOUN fbem-1515 61 18 of of ADP fbem-1515 61 19 examples example NOUN fbem-1515 61 20 ) ) PUNCT fbem-1515 61 21 considering consider VERB fbem-1515 61 22 the the DET fbem-1515 61 23 number number NOUN fbem-1515 61 24 of of ADP fbem-1515 61 25 features feature NOUN fbem-1515 61 26 . . PUNCT fbem-1515 62 1 our our PRON fbem-1515 62 2 next next ADJ fbem-1515 62 3 attempt attempt NOUN fbem-1515 62 4 is be AUX fbem-1515 62 5 directly directly ADV fbem-1515 62 6 building build VERB fbem-1515 62 7 a a DET fbem-1515 62 8 decision decision NOUN fbem-1515 62 9 tree tree NOUN fbem-1515 62 10 model model NOUN fbem-1515 62 11 based base VERB fbem-1515 62 12 on on ADP fbem-1515 62 13 train_df train_df PROPN fbem-1515 62 14 ( ( PUNCT fbem-1515 62 15 the the DET fbem-1515 62 16 data datum NOUN fbem-1515 62 17 set set VERB fbem-1515 62 18 from from ADP fbem-1515 62 19 train_cleaned.csv train_cleaned.csv NUM fbem-1515 62 20 ) ) PUNCT fbem-1515 62 21 . . PUNCT fbem-1515 63 1 recall recall VERB fbem-1515 63 2 that that SCONJ fbem-1515 63 3 we we PRON fbem-1515 63 4 have have AUX fbem-1515 63 5 previously previously ADV fbem-1515 63 6 split split VERB fbem-1515 63 7 3 3 NUM fbem-1515 63 8 parts part NOUN fbem-1515 63 9 in in ADP fbem-1515 63 10 train_df train_df PROPN fbem-1515 63 11 : : PUNCT fbem-1515 63 12 ( ( PUNCT fbem-1515 63 13 1 1 X fbem-1515 63 14 ) ) PUNCT fbem-1515 63 15 sub_train sub_train NOUN fbem-1515 63 16 set set VERB fbem-1515 63 17 ( ( PUNCT fbem-1515 63 18 2)validation 2)validation NUM fbem-1515 63 19 set set NOUN fbem-1515 63 20 ( ( PUNCT fbem-1515 63 21 3 3 NUM fbem-1515 63 22 ) ) PUNCT fbem-1515 63 23 test test NOUN fbem-1515 63 24 set set VERB fbem-1515 63 25 . . PUNCT fbem-1515 64 1 we we PRON fbem-1515 64 2 built build VERB fbem-1515 64 3 a a DET fbem-1515 64 4 second second ADJ fbem-1515 64 5 decision decision NOUN fbem-1515 64 6 tree tree NOUN fbem-1515 64 7 model model NOUN fbem-1515 64 8 , , PUNCT fbem-1515 64 9 dt_model dt_model PROPN fbem-1515 64 10 , , PUNCT fbem-1515 64 11 on on ADP fbem-1515 64 12 the the DET fbem-1515 64 13 sub_train sub_train NOUN fbem-1515 64 14 set set VERB fbem-1515 64 15 . . PUNCT fbem-1515 65 1 and and CCONJ fbem-1515 65 2 then then ADV fbem-1515 65 3 we we PRON fbem-1515 65 4 fit fit VERB fbem-1515 65 5 test test NOUN fbem-1515 65 6 set set NOUN fbem-1515 65 7 of of ADP fbem-1515 65 8 train_df train_df PROPN fbem-1515 65 9 into into ADP fbem-1515 65 10 the the DET fbem-1515 65 11 model model NOUN fbem-1515 65 12 and and CCONJ fbem-1515 65 13 got get VERB fbem-1515 65 14 a a DET fbem-1515 65 15 high high ADJ fbem-1515 65 16 result result NOUN fbem-1515 65 17 accuracy accuracy NOUN fbem-1515 65 18 of of ADP fbem-1515 65 19 0.92 0.92 NUM fbem-1515 65 20 . . PUNCT fbem-1515 66 1 figure figure VERB fbem-1515 66 2 7 7 NUM fbem-1515 66 3 . . PUNCT fbem-1515 67 1 the the DET fbem-1515 67 2 result result NOUN fbem-1515 67 3 table table NOUN fbem-1515 67 4 of of ADP fbem-1515 67 5 decision decision NOUN fbem-1515 67 6 tree tree NOUN fbem-1515 67 7 model model NOUN fbem-1515 67 8 directly directly ADV fbem-1515 67 9 built build VERB fbem-1515 67 10 based base VERB fbem-1515 67 11 on on ADP fbem-1515 67 12 train_df train_df PROPN fbem-1515 67 13 then then ADV fbem-1515 67 14 we we PRON fbem-1515 67 15 examined examine VERB fbem-1515 67 16 performance performance NOUN fbem-1515 67 17 in in ADP fbem-1515 67 18 terms term NOUN fbem-1515 67 19 of of ADP fbem-1515 67 20 the the DET fbem-1515 67 21 auc auc NOUN fbem-1515 67 22 , , PUNCT fbem-1515 67 23 it it PRON fbem-1515 67 24 ’s ’ VERB fbem-1515 67 25 around around ADV fbem-1515 67 26 0.51 0.51 NUM fbem-1515 67 27 . . PUNCT fbem-1515 68 1 figure figure NOUN fbem-1515 68 2 8 8 NUM fbem-1515 68 3 . . PUNCT fbem-1515 69 1 the the DET fbem-1515 69 2 final final ADJ fbem-1515 69 3 auc auc NOUN fbem-1515 69 4 of of ADP fbem-1515 69 5 decision decision NOUN fbem-1515 69 6 tree tree NOUN fbem-1515 69 7 model model NOUN fbem-1515 69 8 directly directly ADV fbem-1515 69 9 built build VERB fbem-1515 69 10 based base VERB fbem-1515 69 11 on on ADP fbem-1515 69 12 train_df train_df X fbem-1515 69 13 considering consider VERB fbem-1515 69 14 the the DET fbem-1515 69 15 high high ADJ fbem-1515 69 16 accuracy accuracy NOUN fbem-1515 69 17 and and CCONJ fbem-1515 69 18 an an DET fbem-1515 69 19 acceptable acceptable ADJ fbem-1515 69 20 auc auc NOUN fbem-1515 69 21 above above ADP fbem-1515 69 22 0.5 0.5 NUM fbem-1515 69 23 , , PUNCT fbem-1515 69 24 we we PRON fbem-1515 69 25 decided decide VERB fbem-1515 69 26 to to PART fbem-1515 69 27 adopt adopt VERB fbem-1515 69 28 dt_model dt_model PROPN fbem-1515 69 29 . . PROPN fbem-1515 69 30 threshold threshold PROPN fbem-1515 69 31 adjustment adjustment NOUN fbem-1515 69 32 based base VERB fbem-1515 69 33 on on ADP fbem-1515 69 34 cost cost NOUN fbem-1515 69 35 - - PUNCT fbem-1515 69 36 benefit benefit NOUN fbem-1515 69 37 analysis analysis NOUN fbem-1515 69 38 145 145 NUM fbem-1515 69 39 after after ADP fbem-1515 69 40 getting get VERB fbem-1515 69 41 the the DET fbem-1515 69 42 model model NOUN fbem-1515 69 43 ( ( PUNCT fbem-1515 69 44 dt_model dt_model NOUN fbem-1515 69 45 ) ) PUNCT fbem-1515 69 46 , , PUNCT fbem-1515 69 47 we we PRON fbem-1515 69 48 adopt adopt VERB fbem-1515 69 49 cost cost NOUN fbem-1515 69 50 and and CCONJ fbem-1515 69 51 benefit benefit VERB fbem-1515 69 52 analysis analysis NOUN fbem-1515 69 53 on on ADP fbem-1515 69 54 our our PRON fbem-1515 69 55 validation validation NOUN fbem-1515 69 56 set set VERB fbem-1515 69 57 to to PART fbem-1515 69 58 find find VERB fbem-1515 69 59 the the DET fbem-1515 69 60 optimal optimal ADJ fbem-1515 69 61 decision decision NOUN fbem-1515 69 62 threshold threshold NOUN fbem-1515 69 63 . . PUNCT fbem-1515 70 1 the the DET fbem-1515 70 2 optimal optimal ADJ fbem-1515 70 3 threshold threshold NOUN fbem-1515 70 4 ( ( PUNCT fbem-1515 70 5 for for ADP fbem-1515 70 6 β€˜ ' PUNCT fbem-1515 70 7 0 0 NUM fbem-1515 70 8 ’ ' PUNCT fbem-1515 70 9 prediction prediction NOUN fbem-1515 70 10 ) ) PUNCT fbem-1515 70 11 is be AUX fbem-1515 70 12 found find VERB fbem-1515 70 13 to to PART fbem-1515 70 14 be be AUX fbem-1515 70 15 0.9 0.9 NUM fbem-1515 70 16 . . PUNCT fbem-1515 71 1 the the DET fbem-1515 71 2 β€˜ ' PUNCT fbem-1515 71 3 cost cost NOUN fbem-1515 71 4 curve curve NOUN fbem-1515 71 5 ’ ' PUNCT fbem-1515 71 6 and and CCONJ fbem-1515 71 7 β€˜ ' PUNCT fbem-1515 71 8 cost cost NOUN fbem-1515 71 9 comparison comparison NOUN fbem-1515 71 10 ’ ' PUNCT fbem-1515 71 11 are be AUX fbem-1515 71 12 shown show VERB fbem-1515 71 13 below below ADP fbem-1515 71 14 : : PUNCT fbem-1515 71 15 figure figure VERB fbem-1515 71 16 9 9 NUM fbem-1515 71 17 . . PUNCT fbem-1515 71 18 cost cost NOUN fbem-1515 71 19 graph graph NOUN fbem-1515 71 20 with with ADP fbem-1515 71 21 different different ADJ fbem-1515 71 22 threshold threshold NOUN fbem-1515 71 23 we we PRON fbem-1515 71 24 then then ADV fbem-1515 71 25 adjusted adjust VERB fbem-1515 71 26 the the DET fbem-1515 71 27 decision decision NOUN fbem-1515 71 28 threshold threshold NOUN fbem-1515 71 29 of of ADP fbem-1515 71 30 dt_model dt_model NOUN fbem-1515 71 31 to to ADP fbem-1515 71 32 0.93 0.93 NUM fbem-1515 71 33 . . PUNCT fbem-1515 72 1 figure figure NOUN fbem-1515 72 2 10 10 NUM fbem-1515 72 3 . . PUNCT fbem-1515 73 1 cost cost NOUN fbem-1515 73 2 comparison comparison NOUN fbem-1515 73 3 table table NOUN fbem-1515 73 4 between between ADP fbem-1515 73 5 all all PRON fbem-1515 73 6 predicted predict VERB fbem-1515 73 7 by by ADP fbem-1515 73 8 majority majority NOUN fbem-1515 73 9 and and CCONJ fbem-1515 73 10 by by ADP fbem-1515 73 11 our our PRON fbem-1515 73 12 0.9 0.9 NUM fbem-1515 73 13 ideal ideal NOUN fbem-1515 73 14 threshold threshold VERB fbem-1515 73 15 the the DET fbem-1515 73 16 cost cost NOUN fbem-1515 73 17 incurred incur VERB fbem-1515 73 18 by by ADP fbem-1515 73 19 a a DET fbem-1515 73 20 majority majority NOUN fbem-1515 73 21 classifier classifier NOUN fbem-1515 73 22 ( ( PUNCT fbem-1515 73 23 all all PRON fbem-1515 73 24 predict predict VERB fbem-1515 73 25 to to PART fbem-1515 73 26 be be AUX fbem-1515 73 27 non non ADJ fbem-1515 73 28 - - NOUN fbem-1515 73 29 default default NOUN fbem-1515 73 30 ) ) PUNCT fbem-1515 73 31 is be AUX fbem-1515 73 32 about about ADV fbem-1515 73 33 43.9 43.9 NUM fbem-1515 73 34 million million NUM fbem-1515 73 35 ( ( PUNCT fbem-1515 73 36 β€œ " PUNCT fbem-1515 73 37 total total ADJ fbem-1515 73 38 cost cost NOUN fbem-1515 73 39 ” " PUNCT fbem-1515 73 40 ) ) PUNCT fbem-1515 73 41 . . PUNCT fbem-1515 74 1 the the DET fbem-1515 74 2 cost cost NOUN fbem-1515 74 3 decreased decrease VERB fbem-1515 74 4 by by ADP fbem-1515 74 5 23.95 23.95 NUM fbem-1515 74 6 % % NOUN fbem-1515 74 7 after after ADP fbem-1515 74 8 the the DET fbem-1515 74 9 prediction prediction NOUN fbem-1515 74 10 with with ADP fbem-1515 74 11 default default NOUN fbem-1515 74 12 threshold threshold NOUN fbem-1515 74 13 of of ADP fbem-1515 74 14 0.93 0.93 NUM fbem-1515 74 15 . . PUNCT fbem-1515 75 1 evaluation evaluation NOUN fbem-1515 75 2 we we PRON fbem-1515 75 3 then then ADV fbem-1515 75 4 checked check VERB fbem-1515 75 5 the the DET fbem-1515 75 6 optimal optimal ADJ fbem-1515 75 7 model model NOUN fbem-1515 75 8 with with ADP fbem-1515 75 9 the the DET fbem-1515 75 10 decision decision NOUN fbem-1515 75 11 threshold threshold NOUN fbem-1515 75 12 adjusted adjust VERB fbem-1515 75 13 on on ADP fbem-1515 75 14 the the DET fbem-1515 75 15 test test NOUN fbem-1515 75 16 set set NOUN fbem-1515 75 17 of of ADP fbem-1515 75 18 train_df train_df PROPN fbem-1515 75 19 to to PART fbem-1515 75 20 get get VERB fbem-1515 75 21 the the DET fbem-1515 75 22 results result NOUN fbem-1515 75 23 as as SCONJ fbem-1515 75 24 follows follow VERB fbem-1515 75 25 : : PUNCT fbem-1515 75 26 accuracy accuracy NOUN fbem-1515 75 27 : : PUNCT fbem-1515 75 28 0.6861 0.6861 NUM fbem-1515 75 29 cost cost NOUN fbem-1515 75 30 reduction reduction NOUN fbem-1515 75 31 : : PUNCT fbem-1515 75 32 22.65 22.65 NUM fbem-1515 75 33 % % NOUN fbem-1515 75 34 reduction reduction NOUN fbem-1515 75 35 compared compare VERB fbem-1515 75 36 to to ADP fbem-1515 75 37 the the DET fbem-1515 75 38 majority majority NOUN fbem-1515 75 39 classifier classifier NOUN fbem-1515 75 40 figure figure VERB fbem-1515 75 41 11 11 NUM fbem-1515 75 42 . . PUNCT fbem-1515 76 1 the the DET fbem-1515 76 2 confusion confusion NOUN fbem-1515 76 3 matrix matrix NOUN fbem-1515 76 4 of of ADP fbem-1515 76 5 test test NOUN fbem-1515 76 6 set set VERB fbem-1515 76 7 based base VERB fbem-1515 76 8 on on ADP fbem-1515 76 9 dt_model dt_model ADJ fbem-1515 76 10 figure figure NOUN fbem-1515 76 11 12 12 NUM fbem-1515 76 12 . . PUNCT fbem-1515 77 1 the the DET fbem-1515 77 2 reduced reduce VERB fbem-1515 77 3 cost cost NOUN fbem-1515 77 4 result result NOUN fbem-1515 77 5 of of ADP fbem-1515 77 6 test test NOUN fbem-1515 77 7 set set VERB fbem-1515 77 8 based base VERB fbem-1515 77 9 on on ADP fbem-1515 77 10 dt_model dt_model ADJ fbem-1515 77 11 conclusion conclusion NOUN fbem-1515 77 12 regarding regard VERB fbem-1515 77 13 the the DET fbem-1515 77 14 2 2 NUM fbem-1515 77 15 decision decision NOUN fbem-1515 77 16 models model NOUN fbem-1515 77 17 we we PRON fbem-1515 77 18 built build VERB fbem-1515 77 19 , , PUNCT fbem-1515 77 20 dt_model_1 dt_model_1 PROPN fbem-1515 77 21 ( ( PUNCT fbem-1515 77 22 nondefault nondefault VERB fbem-1515 77 23 na na PART fbem-1515 77 24 dropped drop VERB fbem-1515 77 25 ) ) PUNCT fbem-1515 77 26 and and CCONJ fbem-1515 77 27 dt_model dt_model NOUN fbem-1515 77 28 , , PUNCT fbem-1515 77 29 though though SCONJ fbem-1515 77 30 we we PRON fbem-1515 77 31 eventually eventually ADV fbem-1515 77 32 decided decide VERB fbem-1515 77 33 to to PART fbem-1515 77 34 adopt adopt VERB fbem-1515 77 35 dt_model dt_model VERB fbem-1515 77 36 , , PUNCT fbem-1515 77 37 we we PRON fbem-1515 77 38 still still ADV fbem-1515 77 39 think think VERB fbem-1515 77 40 that that SCONJ fbem-1515 77 41 the the DET fbem-1515 77 42 feature feature NOUN fbem-1515 77 43 ranking ranking NOUN fbem-1515 77 44 of of ADP fbem-1515 77 45 dt_model_1 dt_model_1 NOUN fbem-1515 77 46 may may AUX fbem-1515 77 47 still still ADV fbem-1515 77 48 give give VERB fbem-1515 77 49 us we PRON fbem-1515 77 50 some some DET fbem-1515 77 51 insights insight NOUN fbem-1515 77 52 of of ADP fbem-1515 77 53 the the DET fbem-1515 77 54 identification identification NOUN fbem-1515 77 55 of of ADP fbem-1515 77 56 default default NOUN fbem-1515 77 57 . . PUNCT fbem-1515 78 1 and and CCONJ fbem-1515 78 2 for for ADP fbem-1515 78 3 dt_model dt_model NOUN fbem-1515 78 4 , , PUNCT fbem-1515 78 5 though though SCONJ fbem-1515 78 6 the the DET fbem-1515 78 7 final final ADJ fbem-1515 78 8 model model NOUN fbem-1515 78 9 has have VERB fbem-1515 78 10 a a DET fbem-1515 78 11 relatively relatively ADV fbem-1515 78 12 okay okay ADJ fbem-1515 78 13 accuracy accuracy NOUN fbem-1515 78 14 ( ( PUNCT fbem-1515 78 15 0.6861 0.6861 NUM fbem-1515 78 16 ) ) PUNCT fbem-1515 78 17 and and CCONJ fbem-1515 78 18 cost cost NOUN fbem-1515 78 19 reduction reduction NOUN fbem-1515 78 20 ( ( PUNCT fbem-1515 78 21 22.65 22.65 NUM fbem-1515 78 22 % % NOUN fbem-1515 78 23 ) ) PUNCT fbem-1515 78 24 , , PUNCT fbem-1515 78 25 the the DET fbem-1515 78 26 auc auc NOUN fbem-1515 78 27 is be AUX fbem-1515 78 28 still still ADV fbem-1515 78 29 unideal unideal ADJ fbem-1515 78 30 , , PUNCT fbem-1515 78 31 being be AUX fbem-1515 78 32 just just ADV fbem-1515 78 33 slightly slightly ADV fbem-1515 78 34 above above ADP fbem-1515 78 35 0.5 0.5 NUM fbem-1515 78 36 . . PUNCT fbem-1515 79 1 it it PRON fbem-1515 79 2 still still ADV fbem-1515 79 3 needs need VERB fbem-1515 79 4 improvement improvement NOUN fbem-1515 79 5 . . PUNCT fbem-1515 80 1 4.3 4.3 NUM fbem-1515 80 2 . . PUNCT fbem-1515 80 3 logistic logistic ADJ fbem-1515 80 4 model model NOUN fbem-1515 80 5 data data PROPN fbem-1515 80 6 source source NOUN fbem-1515 80 7 : : PUNCT fbem-1515 80 8 train_cleaned.csv(β€œtrain train_cleaned.csv(β€œtrain NOUN fbem-1515 80 9 ” " PUNCT fbem-1515 80 10 ) ) PUNCT fbem-1515 80 11 , , PUNCT fbem-1515 80 12 train_1_cleaned.csv train_1_cleaned.csv NUM fbem-1515 80 13 ( ( PUNCT fbem-1515 80 14 β€œ " PUNCT fbem-1515 80 15 train_1 train_1 PROPN fbem-1515 80 16 ” " PUNCT fbem-1515 80 17 ) ) PUNCT fbem-1515 80 18 hyper hyper NOUN fbem-1515 80 19 - - ADJ fbem-1515 80 20 tunning tunning NOUN fbem-1515 80 21 : : PUNCT fbem-1515 80 22 after after ADP fbem-1515 80 23 searching search VERB fbem-1515 80 24 for for ADP fbem-1515 80 25 related related ADJ fbem-1515 80 26 research research NOUN fbem-1515 80 27 about about ADP fbem-1515 80 28 hyper hyper ADJ fbem-1515 80 29 parameters parameter NOUN fbem-1515 80 30 in in ADP fbem-1515 80 31 logistic logistic ADJ fbem-1515 80 32 model model NOUN fbem-1515 80 33 , , PUNCT fbem-1515 80 34 we we PRON fbem-1515 80 35 found find VERB fbem-1515 80 36 that that SCONJ fbem-1515 80 37 the the DET fbem-1515 80 38 hyper hyper ADJ fbem-1515 80 39 parameters parameter NOUN fbem-1515 80 40 that that PRON fbem-1515 80 41 have have VERB fbem-1515 80 42 the the DET fbem-1515 80 43 highest high ADJ fbem-1515 80 44 relationship relationship NOUN fbem-1515 80 45 and and CCONJ fbem-1515 80 46 biggest big ADJ fbem-1515 80 47 influence influence NOUN fbem-1515 80 48 towards towards ADP fbem-1515 80 49 logistic logistic ADJ fbem-1515 80 50 model model NOUN fbem-1515 80 51 is be AUX fbem-1515 80 52 the the DET fbem-1515 80 53 penalty penalty NOUN fbem-1515 80 54 type type NOUN fbem-1515 80 55 ( ( PUNCT fbem-1515 80 56 whether whether SCONJ fbem-1515 80 57 β€œ " PUNCT fbem-1515 80 58 l1 l1 PROPN fbem-1515 80 59 ” " PUNCT fbem-1515 80 60 or or CCONJ fbem-1515 80 61 β€œ " PUNCT fbem-1515 80 62 l2 l2 NOUN fbem-1515 80 63 ” " PUNCT fbem-1515 80 64 ) ) PUNCT fbem-1515 80 65 and and CCONJ fbem-1515 80 66 the the DET fbem-1515 80 67 value value NOUN fbem-1515 80 68 of of ADP fbem-1515 80 69 c c PROPN fbem-1515 80 70 ( ( PUNCT fbem-1515 80 71 which which PRON fbem-1515 80 72 determines determine VERB fbem-1515 80 73 the the DET fbem-1515 80 74 strength strength NOUN fbem-1515 80 75 of of ADP fbem-1515 80 76 penalty penalty NOUN fbem-1515 80 77 ) ) PUNCT fbem-1515 80 78 . . PUNCT fbem-1515 81 1 we we PRON fbem-1515 81 2 confine confine VERB fbem-1515 81 3 those those DET fbem-1515 81 4 2 2 NUM fbem-1515 81 5 parameters parameter NOUN fbem-1515 81 6 into into ADP fbem-1515 81 7 the the DET fbem-1515 81 8 range range NOUN fbem-1515 81 9 of of ADP fbem-1515 81 10 penalty penalty NOUN fbem-1515 81 11 = = PUNCT fbem-1515 82 1 [ [ X fbem-1515 82 2 ' ' PUNCT fbem-1515 82 3 l1 l1 PROPN fbem-1515 82 4 ' ' PUNCT fbem-1515 82 5 , , PUNCT fbem-1515 82 6 ' ' PUNCT fbem-1515 82 7 l2 l2 NOUN fbem-1515 82 8 ' ' PUNCT fbem-1515 82 9 ] ] PUNCT fbem-1515 82 10 ” " PUNCT fbem-1515 82 11 , , PUNCT fbem-1515 82 12 and and CCONJ fbem-1515 82 13 c c X fbem-1515 82 14 = = PUNCT fbem-1515 83 1 [ [ X fbem-1515 83 2 100 100 NUM fbem-1515 83 3 , , PUNCT fbem-1515 83 4 10 10 NUM fbem-1515 83 5 , , PUNCT fbem-1515 83 6 1.0 1.0 NUM fbem-1515 83 7 , , PUNCT fbem-1515 83 8 0.1 0.1 NUM fbem-1515 83 9 , , PUNCT fbem-1515 83 10 0.01 0.01 NUM fbem-1515 83 11 ] ] PUNCT fbem-1515 83 12 , , PUNCT fbem-1515 83 13 then then ADV fbem-1515 83 14 import import VERB fbem-1515 83 15 the the DET fbem-1515 83 16 library library NOUN fbem-1515 83 17 – – PUNCT fbem-1515 83 18 gridsearchcv gridsearchcv VERB fbem-1515 83 19 to to PART fbem-1515 83 20 help help VERB fbem-1515 83 21 us we PRON fbem-1515 83 22 try try VERB fbem-1515 83 23 different different ADJ fbem-1515 83 24 combinations combination NOUN fbem-1515 83 25 of of ADP fbem-1515 83 26 hyper hyper ADJ fbem-1515 83 27 parameters parameter NOUN fbem-1515 83 28 one one NUM fbem-1515 83 29 by by ADP fbem-1515 83 30 one one NUM fbem-1515 83 31 . . PUNCT fbem-1515 84 1 finally finally ADV fbem-1515 84 2 , , PUNCT fbem-1515 84 3 we we PRON fbem-1515 84 4 got get VERB fbem-1515 84 5 the the DET fbem-1515 84 6 best good ADJ fbem-1515 84 7 result result NOUN fbem-1515 84 8 of of ADP fbem-1515 84 9 those those DET fbem-1515 84 10 values value NOUN fbem-1515 84 11 and and CCONJ fbem-1515 84 12 redefine redefine VERB fbem-1515 84 13 our our PRON fbem-1515 84 14 model model NOUN fbem-1515 84 15 based base VERB fbem-1515 84 16 on on ADP fbem-1515 84 17 them they PRON fbem-1515 84 18 , , PUNCT fbem-1515 84 19 which which PRON fbem-1515 84 20 are be AUX fbem-1515 84 21 { { PUNCT fbem-1515 84 22 ' ' PUNCT fbem-1515 84 23 c c NOUN fbem-1515 84 24 ' ' PUNCT fbem-1515 84 25 : : PUNCT fbem-1515 84 26 100 100 NUM fbem-1515 84 27 , , PUNCT fbem-1515 84 28 ' ' PUNCT fbem-1515 84 29 penalty penalty NOUN fbem-1515 84 30 ' ' PUNCT fbem-1515 84 31 : : PUNCT fbem-1515 84 32 ' ' PUNCT fbem-1515 84 33 l1 l1 PROPN fbem-1515 84 34 ' ' PUNCT fbem-1515 84 35 } } PUNCT fbem-1515 84 36 and and CCONJ fbem-1515 84 37 { { PUNCT fbem-1515 84 38 ' ' PUNCT fbem-1515 84 39 c c X fbem-1515 84 40 ' ' PUNCT fbem-1515 84 41 : : PUNCT fbem-1515 84 42 0.01 0.01 NUM fbem-1515 84 43 , , PUNCT fbem-1515 84 44 ' ' PUNCT fbem-1515 84 45 penalty penalty NOUN fbem-1515 84 46 ' ' PUNCT fbem-1515 84 47 : : PUNCT fbem-1515 84 48 ' ' PUNCT fbem-1515 84 49 l1 l1 PROPN fbem-1515 84 50 ' ' PUNCT fbem-1515 84 51 } } PUNCT fbem-1515 84 52 respectively respectively ADV fbem-1515 84 53 in in ADP fbem-1515 84 54 our our PRON fbem-1515 84 55 2 2 NUM fbem-1515 84 56 trials trial NOUN fbem-1515 84 57 . . PUNCT fbem-1515 85 1 model model NOUN fbem-1515 85 2 training training NOUN fbem-1515 85 3 and and CCONJ fbem-1515 85 4 threshold threshold NOUN fbem-1515 85 5 selection selection NOUN fbem-1515 85 6 we we PRON fbem-1515 85 7 have have AUX fbem-1515 85 8 built build VERB fbem-1515 85 9 our our PRON fbem-1515 85 10 model model NOUN fbem-1515 85 11 mainly mainly ADV fbem-1515 85 12 based base VERB fbem-1515 85 13 on on ADP fbem-1515 85 14 two two NUM fbem-1515 85 15 data data NOUN fbem-1515 85 16 sets set NOUN fbem-1515 85 17 : : PUNCT fbem-1515 85 18 train_1_cleaned train_1_cleane VERB fbem-1515 85 19 ( ( PUNCT fbem-1515 85 20 train_1_df train_1_df PROPN fbem-1515 85 21 ) ) PUNCT fbem-1515 85 22 and and CCONJ fbem-1515 85 23 train_cleaned train_cleane VERB fbem-1515 85 24 ( ( PUNCT fbem-1515 85 25 train_df train_df PROPN fbem-1515 85 26 ) ) PUNCT fbem-1515 85 27 , , PUNCT fbem-1515 85 28 and and CCONJ fbem-1515 85 29 would would AUX fbem-1515 85 30 decide decide VERB fbem-1515 85 31 which which DET fbem-1515 85 32 one one NOUN fbem-1515 85 33 to to PART fbem-1515 85 34 use use VERB fbem-1515 85 35 according accord VERB fbem-1515 85 36 to to ADP fbem-1515 85 37 their their PRON fbem-1515 85 38 performance performance NOUN fbem-1515 85 39 evaluated evaluate VERB fbem-1515 85 40 on on ADP fbem-1515 85 41 the the DET fbem-1515 85 42 same same ADJ fbem-1515 85 43 test test NOUN fbem-1515 85 44 set set NOUN fbem-1515 85 45 . . PUNCT fbem-1515 86 1 then then ADV fbem-1515 86 2 we we PRON fbem-1515 86 3 build build VERB fbem-1515 86 4 a a DET fbem-1515 86 5 logistic logistic ADJ fbem-1515 86 6 model model NOUN fbem-1515 86 7 with with ADP fbem-1515 86 8 the the DET fbem-1515 86 9 train_1_cleaned train_1_cleane VERB fbem-1515 86 10 set set NOUN fbem-1515 86 11 and and CCONJ fbem-1515 86 12 use use VERB fbem-1515 86 13 this this DET fbem-1515 86 14 model model NOUN fbem-1515 86 15 to to PART fbem-1515 86 16 fit fit VERB fbem-1515 86 17 into into ADP fbem-1515 86 18 all all DET fbem-1515 86 19 train train NOUN fbem-1515 86 20 data datum NOUN fbem-1515 86 21 ( ( PUNCT fbem-1515 86 22 without without ADP fbem-1515 86 23 dropping drop VERB fbem-1515 86 24 those those DET fbem-1515 86 25 non non ADJ fbem-1515 86 26 - - ADJ fbem-1515 86 27 default default ADJ fbem-1515 86 28 items item NOUN fbem-1515 86 29 ) ) PUNCT fbem-1515 86 30 to to PART fbem-1515 86 31 see see VERB fbem-1515 86 32 the the DET fbem-1515 86 33 accuracy accuracy NOUN fbem-1515 86 34 . . PUNCT fbem-1515 87 1 and and CCONJ fbem-1515 87 2 we we PRON fbem-1515 87 3 found find VERB fbem-1515 87 4 that that SCONJ fbem-1515 87 5 the the DET fbem-1515 87 6 accuracy accuracy NOUN fbem-1515 87 7 of of ADP fbem-1515 87 8 this this DET fbem-1515 87 9 model model NOUN fbem-1515 87 10 is be AUX fbem-1515 87 11 rather rather ADV fbem-1515 87 12 low low ADJ fbem-1515 87 13 in in ADP fbem-1515 87 14 the the DET fbem-1515 87 15 whole whole ADJ fbem-1515 87 16 train train NOUN fbem-1515 87 17 data datum NOUN fbem-1515 87 18 ( ( PUNCT fbem-1515 87 19 exhibit exhibit NOUN fbem-1515 87 20 5 5 NUM fbem-1515 87 21 ) ) PUNCT fbem-1515 87 22 , , PUNCT fbem-1515 87 23 although although SCONJ fbem-1515 87 24 it it PRON fbem-1515 87 25 performed perform VERB fbem-1515 87 26 well well ADV fbem-1515 87 27 in in ADP fbem-1515 87 28 train_1_cleaned train_1_cleaned ADJ fbem-1515 87 29 , , PUNCT fbem-1515 87 30 so so CCONJ fbem-1515 87 31 our our PRON fbem-1515 87 32 first first ADJ fbem-1515 87 33 attempt attempt NOUN fbem-1515 87 34 failed fail VERB fbem-1515 87 35 . . PUNCT fbem-1515 88 1 figure figure VERB fbem-1515 88 2 13 13 NUM fbem-1515 88 3 . . PUNCT fbem-1515 89 1 the the DET fbem-1515 89 2 accuracy accuracy NOUN fbem-1515 89 3 of of ADP fbem-1515 89 4 model model NOUN fbem-1515 89 5 1 1 NUM fbem-1515 89 6 in in ADP fbem-1515 89 7 predicting predict VERB fbem-1515 89 8 whole whole ADJ fbem-1515 89 9 train train NOUN fbem-1515 89 10 data datum NOUN fbem-1515 89 11 our our PRON fbem-1515 89 12 next next ADJ fbem-1515 89 13 attempt attempt NOUN fbem-1515 89 14 is be AUX fbem-1515 89 15 directly directly ADV fbem-1515 89 16 building build VERB fbem-1515 89 17 a a DET fbem-1515 89 18 model model NOUN fbem-1515 89 19 based base VERB fbem-1515 89 20 on on ADP fbem-1515 89 21 the the DET fbem-1515 89 22 whole whole ADJ fbem-1515 89 23 train train NOUN fbem-1515 89 24 data datum NOUN fbem-1515 89 25 set set VERB fbem-1515 89 26 that that PRON fbem-1515 89 27 has have AUX fbem-1515 89 28 been be AUX fbem-1515 89 29 processed process VERB fbem-1515 89 30 , , PUNCT fbem-1515 89 31 which which PRON fbem-1515 89 32 is be AUX fbem-1515 89 33 train_cleaned.csv train_cleaned.csv NUM fbem-1515 89 34 . . PUNCT fbem-1515 90 1 we we PRON fbem-1515 90 2 split split VERB fbem-1515 90 3 this this DET fbem-1515 90 4 data datum NOUN fbem-1515 90 5 set set VERB fbem-1515 90 6 into into ADP fbem-1515 90 7 traino traino NOUN fbem-1515 90 8 set set NOUN fbem-1515 90 9 and and CCONJ fbem-1515 90 10 test test NOUN fbem-1515 90 11 set set NOUN fbem-1515 90 12 , , PUNCT fbem-1515 90 13 then then ADV fbem-1515 90 14 further far ADV fbem-1515 90 15 split split ADJ fbem-1515 90 16 traino traino NOUN fbem-1515 90 17 set set VERB fbem-1515 90 18 into into ADP fbem-1515 90 19 train train NOUN fbem-1515 90 20 set set NOUN fbem-1515 90 21 and and CCONJ fbem-1515 90 22 validation validation NOUN fbem-1515 90 23 set(which set(which PRON fbem-1515 90 24 we we PRON fbem-1515 90 25 will will AUX fbem-1515 90 26 use use VERB fbem-1515 90 27 later later ADV fbem-1515 90 28 for for ADP fbem-1515 90 29 cost cost NOUN fbem-1515 90 30 analysis analysis NOUN fbem-1515 90 31 ) ) PUNCT fbem-1515 90 32 , , PUNCT fbem-1515 90 33 then then ADV fbem-1515 90 34 build build VERB fbem-1515 90 35 model model NOUN fbem-1515 90 36 based base VERB fbem-1515 90 37 on on ADP fbem-1515 90 38 train train NOUN fbem-1515 90 39 set set NOUN fbem-1515 90 40 and and CCONJ fbem-1515 90 41 valuation valuation NOUN fbem-1515 90 42 based base VERB fbem-1515 90 43 on on ADP fbem-1515 90 44 test test NOUN fbem-1515 90 45 set set VERB fbem-1515 90 46 . . PUNCT fbem-1515 91 1 we we PRON fbem-1515 91 2 found find VERB fbem-1515 91 3 that that SCONJ fbem-1515 91 4 this this DET fbem-1515 91 5 time time NOUN fbem-1515 91 6 the the DET fbem-1515 91 7 accuracy accuracy NOUN fbem-1515 91 8 of of ADP fbem-1515 91 9 our our PRON fbem-1515 91 10 model model NOUN fbem-1515 91 11 is be AUX fbem-1515 91 12 pretty pretty ADV fbem-1515 91 13 high high ADJ fbem-1515 91 14 ( ( PUNCT fbem-1515 91 15 0.919867 0.919867 NUM fbem-1515 91 16 ) ) PUNCT fbem-1515 91 17 , , PUNCT fbem-1515 91 18 and and CCONJ fbem-1515 91 19 the the DET fbem-1515 91 20 auc auc NOUN fbem-1515 91 21 ( ( PUNCT fbem-1515 91 22 0.500804 0.500804 NUM fbem-1515 91 23 ) ) PUNCT fbem-1515 91 24 also also ADV fbem-1515 91 25 passed pass VERB fbem-1515 91 26 50 50 NUM fbem-1515 91 27 % % NOUN fbem-1515 91 28 , , PUNCT fbem-1515 91 29 which which PRON fbem-1515 91 30 shows show VERB fbem-1515 91 31 the the DET fbem-1515 91 32 usefulness usefulness NOUN fbem-1515 91 33 of of ADP fbem-1515 91 34 our our PRON fbem-1515 91 35 second second ADJ fbem-1515 91 36 trial trial NOUN fbem-1515 91 37 model model NOUN fbem-1515 91 38 . . PUNCT fbem-1515 92 1 threshold threshold NOUN fbem-1515 92 2 selection selection NOUN fbem-1515 92 3 : : PUNCT fbem-1515 92 4 after after ADP fbem-1515 92 5 getting get VERB fbem-1515 92 6 the the DET fbem-1515 92 7 model model NOUN fbem-1515 92 8 , , PUNCT fbem-1515 92 9 we we PRON fbem-1515 92 10 adopt adopt VERB fbem-1515 92 11 cost cost NOUN fbem-1515 92 12 and and CCONJ fbem-1515 92 13 benefit benefit VERB fbem-1515 92 14 analysis analysis NOUN fbem-1515 92 15 on on ADP fbem-1515 92 16 our our PRON fbem-1515 92 17 validation validation NOUN fbem-1515 92 18 set set NOUN fbem-1515 92 19 . . PUNCT fbem-1515 93 1 the the DET fbem-1515 93 2 optimal optimal ADJ fbem-1515 93 3 threshold threshold NOUN fbem-1515 93 4 ( ( PUNCT fbem-1515 93 5 for for ADP fbem-1515 93 6 β€˜ ' PUNCT fbem-1515 93 7 0 0 NUM fbem-1515 93 8 ’ ' PUNCT fbem-1515 93 9 prediction prediction NOUN fbem-1515 93 10 ) ) PUNCT fbem-1515 93 11 is be AUX fbem-1515 93 12 found find VERB fbem-1515 93 13 to to PART fbem-1515 93 14 be be AUX fbem-1515 93 15 0.9 0.9 NUM fbem-1515 93 16 . . PUNCT fbem-1515 94 1 the the DET fbem-1515 94 2 β€˜ ' PUNCT fbem-1515 94 3 cost cost NOUN fbem-1515 94 4 curve curve NOUN fbem-1515 94 5 ’ ' PUNCT fbem-1515 94 6 and and CCONJ fbem-1515 94 7 β€˜ ' PUNCT fbem-1515 94 8 cost cost NOUN fbem-1515 94 9 comparison comparison NOUN fbem-1515 94 10 ’ ' PUNCT fbem-1515 94 11 are be AUX fbem-1515 94 12 shown show VERB fbem-1515 94 13 below below ADP fbem-1515 94 14 : : PUNCT fbem-1515 94 15 146 146 NUM fbem-1515 94 16 figure figure NOUN fbem-1515 94 17 14 14 NUM fbem-1515 94 18 . . PUNCT fbem-1515 95 1 cost cost NOUN fbem-1515 95 2 graph graph NOUN fbem-1515 95 3 with with ADP fbem-1515 95 4 different different ADJ fbem-1515 95 5 threshold threshold NOUN fbem-1515 95 6 & & CCONJ fbem-1515 95 7 cost cost VERB fbem-1515 95 8 comparison comparison NOUN fbem-1515 95 9 table table NOUN fbem-1515 95 10 between between ADP fbem-1515 95 11 all all PRON fbem-1515 95 12 predicted predict VERB fbem-1515 95 13 by by ADP fbem-1515 95 14 majority majority NOUN fbem-1515 95 15 , , PUNCT fbem-1515 95 16 predicted predict VERB fbem-1515 95 17 without without ADP fbem-1515 95 18 threshold threshold NOUN fbem-1515 95 19 adjustment adjustment NOUN fbem-1515 95 20 , , PUNCT fbem-1515 95 21 and and CCONJ fbem-1515 95 22 by by ADP fbem-1515 95 23 our our PRON fbem-1515 95 24 0.9 0.9 NUM fbem-1515 95 25 ideal ideal NOUN fbem-1515 95 26 threshold threshold VERB fbem-1515 95 27 the the DET fbem-1515 95 28 cost cost NOUN fbem-1515 95 29 of of ADP fbem-1515 95 30 validation validation NOUN fbem-1515 95 31 set set NOUN fbem-1515 95 32 incurred incur VERB fbem-1515 95 33 by by ADP fbem-1515 95 34 a a DET fbem-1515 95 35 majority majority NOUN fbem-1515 95 36 classifier classifier NOUN fbem-1515 95 37 ( ( PUNCT fbem-1515 95 38 all all PRON fbem-1515 95 39 predict predict VERB fbem-1515 95 40 to to PART fbem-1515 95 41 be be AUX fbem-1515 95 42 non non ADJ fbem-1515 95 43 - - NOUN fbem-1515 95 44 default default NOUN fbem-1515 95 45 ) ) PUNCT fbem-1515 95 46 is be AUX fbem-1515 95 47 about about ADV fbem-1515 95 48 50.7 50.7 NUM fbem-1515 95 49 million million NUM fbem-1515 95 50 ( ( PUNCT fbem-1515 95 51 β€œ " PUNCT fbem-1515 95 52 total total ADJ fbem-1515 95 53 cost cost NOUN fbem-1515 95 54 ” " PUNCT fbem-1515 95 55 ) ) PUNCT fbem-1515 95 56 . . PUNCT fbem-1515 96 1 the the DET fbem-1515 96 2 cost cost NOUN fbem-1515 96 3 decreased decrease VERB fbem-1515 96 4 by by ADP fbem-1515 96 5 0.2 0.2 NUM fbem-1515 96 6 % % NOUN fbem-1515 96 7 in in ADP fbem-1515 96 8 our our PRON fbem-1515 96 9 model model NOUN fbem-1515 96 10 without without ADP fbem-1515 96 11 threshold threshold NOUN fbem-1515 96 12 adjustment adjustment NOUN fbem-1515 96 13 ( ( PUNCT fbem-1515 96 14 threshold threshold NOUN fbem-1515 96 15 being be AUX fbem-1515 96 16 0.5 0.5 NUM fbem-1515 96 17 ) ) PUNCT fbem-1515 96 18 and and CCONJ fbem-1515 96 19 27.06 27.06 NUM fbem-1515 96 20 % % NOUN fbem-1515 96 21 after after ADP fbem-1515 96 22 the the DET fbem-1515 96 23 prediction prediction NOUN fbem-1515 96 24 with with ADP fbem-1515 96 25 the the DET fbem-1515 96 26 0.9 0.9 NUM fbem-1515 96 27 threshold threshold NOUN fbem-1515 96 28 . . PUNCT fbem-1515 97 1 evaluation evaluation NOUN fbem-1515 97 2 we we PRON fbem-1515 97 3 have have AUX fbem-1515 97 4 applied apply VERB fbem-1515 97 5 the the DET fbem-1515 97 6 model model NOUN fbem-1515 97 7 and and CCONJ fbem-1515 97 8 the the DET fbem-1515 97 9 optimal optimal ADJ fbem-1515 97 10 threshold threshold NOUN fbem-1515 97 11 on on ADP fbem-1515 97 12 the the DET fbem-1515 97 13 test test NOUN fbem-1515 97 14 set set VERB fbem-1515 97 15 and and CCONJ fbem-1515 97 16 get get VERB fbem-1515 97 17 the the DET fbem-1515 97 18 results result NOUN fbem-1515 97 19 as as SCONJ fbem-1515 97 20 follows follow VERB fbem-1515 97 21 : : PUNCT fbem-1515 97 22 figure figure NOUN fbem-1515 97 23 15 15 NUM fbem-1515 97 24 . . PUNCT fbem-1515 98 1 the the DET fbem-1515 98 2 confusion confusion NOUN fbem-1515 98 3 matrix matrix NOUN fbem-1515 98 4 of of ADP fbem-1515 98 5 test test NOUN fbem-1515 98 6 set set VERB fbem-1515 98 7 based base VERB fbem-1515 98 8 on on ADP fbem-1515 98 9 logistic logistic ADJ fbem-1515 98 10 model model NOUN fbem-1515 98 11 figure figure NOUN fbem-1515 98 12 16 16 NUM fbem-1515 98 13 . . PUNCT fbem-1515 99 1 the the DET fbem-1515 99 2 reduced reduce VERB fbem-1515 99 3 cost cost NOUN fbem-1515 99 4 result result NOUN fbem-1515 99 5 of of ADP fbem-1515 99 6 test test NOUN fbem-1515 99 7 set set VERB fbem-1515 99 8 based base VERB fbem-1515 99 9 on on ADP fbem-1515 99 10 logistic logistic ADJ fbem-1515 99 11 model model NOUN fbem-1515 99 12 the the DET fbem-1515 99 13 cost cost NOUN fbem-1515 99 14 is be AUX fbem-1515 99 15 reduced reduce VERB fbem-1515 99 16 by by ADP fbem-1515 99 17 about about ADV fbem-1515 99 18 26.6 26.6 NUM fbem-1515 99 19 % % NOUN fbem-1515 99 20 in in ADP fbem-1515 99 21 test test NOUN fbem-1515 99 22 set set VERB fbem-1515 99 23 after after ADP fbem-1515 99 24 applying apply VERB fbem-1515 99 25 the the DET fbem-1515 99 26 model model NOUN fbem-1515 99 27 . . PUNCT fbem-1515 100 1 conclusion conclusion NOUN fbem-1515 100 2 although although SCONJ fbem-1515 100 3 our our PRON fbem-1515 100 4 logistic logistic ADJ fbem-1515 100 5 model model NOUN fbem-1515 100 6 has have VERB fbem-1515 100 7 pretty pretty ADV fbem-1515 100 8 high high ADJ fbem-1515 100 9 accuracy accuracy NOUN fbem-1515 100 10 after after ADP fbem-1515 100 11 threshold threshold NOUN fbem-1515 100 12 adjustment adjustment NOUN fbem-1515 100 13 ( ( PUNCT fbem-1515 100 14 0.7507 0.7507 NUM fbem-1515 100 15 ) ) PUNCT fbem-1515 100 16 and and CCONJ fbem-1515 100 17 helps help VERB fbem-1515 100 18 reduce reduce VERB fbem-1515 100 19 costs cost NOUN fbem-1515 100 20 by by ADP fbem-1515 100 21 about about ADV fbem-1515 100 22 26.6 26.6 NUM fbem-1515 100 23 % % NOUN fbem-1515 100 24 , , PUNCT fbem-1515 100 25 but but CCONJ fbem-1515 100 26 our our PRON fbem-1515 100 27 auc auc NOUN fbem-1515 100 28 ( ( PUNCT fbem-1515 100 29 0.500804 0.500804 NUM fbem-1515 100 30 ) ) PUNCT fbem-1515 100 31 is be AUX fbem-1515 100 32 still still ADV fbem-1515 100 33 not not PART fbem-1515 100 34 so so ADV fbem-1515 100 35 satisfactory satisfactory ADJ fbem-1515 100 36 as as SCONJ fbem-1515 100 37 we we PRON fbem-1515 100 38 expected expect VERB fbem-1515 100 39 . . PUNCT fbem-1515 101 1 so so ADV fbem-1515 101 2 , , PUNCT fbem-1515 101 3 we we PRON fbem-1515 101 4 kept keep VERB fbem-1515 101 5 trying try VERB fbem-1515 101 6 other other ADJ fbem-1515 101 7 models model NOUN fbem-1515 101 8 to to PART fbem-1515 101 9 find find VERB fbem-1515 101 10 more more ADV fbem-1515 101 11 precise precise ADJ fbem-1515 101 12 results result NOUN fbem-1515 101 13 . . PUNCT fbem-1515 102 1 4.4 4.4 NUM fbem-1515 102 2 . . PUNCT fbem-1515 103 1 knn knn PROPN fbem-1515 103 2 model model PROPN fbem-1515 103 3 data data PROPN fbem-1515 103 4 source source NOUN fbem-1515 103 5 : : PUNCT fbem-1515 103 6 train_cleaned.csv(β€œtrain train_cleaned.csv(β€œtrain NOUN fbem-1515 103 7 ” " PUNCT fbem-1515 103 8 ) ) PUNCT fbem-1515 103 9 , , PUNCT fbem-1515 103 10 train_1_cleaned.csv train_1_cleaned.csv NUM fbem-1515 103 11 ( ( PUNCT fbem-1515 103 12 β€œ " PUNCT fbem-1515 103 13 train_1 train_1 PROPN fbem-1515 103 14 ” " PUNCT fbem-1515 103 15 ) ) PUNCT fbem-1515 103 16 hyper hyper NOUN fbem-1515 103 17 - - ADJ fbem-1515 103 18 tunning tunning NOUN fbem-1515 103 19 : : PUNCT fbem-1515 103 20 in in ADP fbem-1515 103 21 the the DET fbem-1515 103 22 model model NOUN fbem-1515 103 23 , , PUNCT fbem-1515 103 24 we we PRON fbem-1515 103 25 have have AUX fbem-1515 103 26 used use VERB fbem-1515 103 27 2 2 NUM fbem-1515 103 28 parameters parameter NOUN fbem-1515 103 29 [ [ X fbem-1515 103 30 neighbor neighbor NOUN fbem-1515 103 31 count count NOUN fbem-1515 103 32 ( ( PUNCT fbem-1515 103 33 k k NOUN fbem-1515 103 34 ) ) PUNCT fbem-1515 103 35 , , PUNCT fbem-1515 103 36 distance distance NOUN fbem-1515 103 37 type type NOUN fbem-1515 103 38 ( ( PUNCT fbem-1515 103 39 p p NOUN fbem-1515 103 40 ) ) PUNCT fbem-1515 103 41 ] ] PUNCT fbem-1515 103 42 to to PART fbem-1515 103 43 assign assign VERB fbem-1515 103 44 weights weight NOUN fbem-1515 103 45 to to ADP fbem-1515 103 46 neighbors neighbor NOUN fbem-1515 103 47 by by ADP fbem-1515 103 48 distance distance NOUN fbem-1515 103 49 . . PUNCT fbem-1515 104 1 as as SCONJ fbem-1515 104 2 we we PRON fbem-1515 104 3 care care VERB fbem-1515 104 4 more more ADJ fbem-1515 104 5 about about ADP fbem-1515 104 6 positive positive ADJ fbem-1515 104 7 labels label NOUN fbem-1515 104 8 , , PUNCT fbem-1515 104 9 we we PRON fbem-1515 104 10 use use VERB fbem-1515 104 11 auc auc NOUN fbem-1515 104 12 as as ADP fbem-1515 104 13 the the DET fbem-1515 104 14 performance performance NOUN fbem-1515 104 15 index index NOUN fbem-1515 104 16 in in ADP fbem-1515 104 17 the the DET fbem-1515 104 18 tunning tunning NOUN fbem-1515 104 19 stage stage NOUN fbem-1515 104 20 . . PUNCT fbem-1515 105 1 for for ADP fbem-1515 105 2 the the DET fbem-1515 105 3 train_1 train_1 PROPN fbem-1515 105 4 data datum NOUN fbem-1515 105 5 , , PUNCT fbem-1515 105 6 15 15 NUM fbem-1515 105 7 combinations combination NOUN fbem-1515 105 8 of of ADP fbem-1515 105 9 the the DET fbem-1515 105 10 two two NUM fbem-1515 105 11 parameters parameter NOUN fbem-1515 105 12 are be AUX fbem-1515 105 13 used use VERB fbem-1515 105 14 to to PART fbem-1515 105 15 tune tune VERB fbem-1515 105 16 the the DET fbem-1515 105 17 optimal optimal ADJ fbem-1515 105 18 parameters parameter NOUN fbem-1515 105 19 with with ADP fbem-1515 105 20 cross cross NOUN fbem-1515 105 21 - - NOUN fbem-1515 105 22 validation validation NOUN fbem-1515 105 23 . . PUNCT fbem-1515 106 1 for for ADP fbem-1515 106 2 train train NOUN fbem-1515 106 3 data datum NOUN fbem-1515 106 4 , , PUNCT fbem-1515 106 5 11 11 NUM fbem-1515 106 6 combinations combination NOUN fbem-1515 106 7 are be AUX fbem-1515 106 8 used use VERB fbem-1515 106 9 to to PART fbem-1515 106 10 tune tune VERB fbem-1515 106 11 the the DET fbem-1515 106 12 model model NOUN fbem-1515 106 13 . . PUNCT fbem-1515 107 1 among among ADP fbem-1515 107 2 the the DET fbem-1515 107 3 combinations combination NOUN fbem-1515 107 4 , , PUNCT fbem-1515 107 5 we we PRON fbem-1515 107 6 figure figure VERB fbem-1515 107 7 out out ADP fbem-1515 107 8 that that SCONJ fbem-1515 107 9 the the DET fbem-1515 107 10 relative relative ADJ fbem-1515 107 11 auc auc NOUN fbem-1515 107 12 maximum maximum NOUN fbem-1515 107 13 is be AUX fbem-1515 107 14 achieved achieve VERB fbem-1515 107 15 when when SCONJ fbem-1515 107 16 k k PROPN fbem-1515 107 17 = = PUNCT fbem-1515 107 18 33/35 33/35 PROPN fbem-1515 107 19 and and CCONJ fbem-1515 107 20 p p X fbem-1515 107 21 = = NOUN fbem-1515 107 22 1 1 NUM fbem-1515 107 23 for for ADP fbem-1515 107 24 train_1 train_1 NOUN fbem-1515 107 25 ; ; PUNCT fbem-1515 107 26 and and CCONJ fbem-1515 107 27 when when SCONJ fbem-1515 107 28 k=71 k=71 PROPN fbem-1515 107 29 and and CCONJ fbem-1515 107 30 p=1 p=1 NOUN fbem-1515 107 31 for for ADP fbem-1515 107 32 train train NOUN fbem-1515 107 33 . . PUNCT fbem-1515 108 1 model model NOUN fbem-1515 108 2 training training NOUN fbem-1515 108 3 and and CCONJ fbem-1515 108 4 threshold threshold NOUN fbem-1515 108 5 selection selection NOUN fbem-1515 108 6 we we PRON fbem-1515 108 7 have have AUX fbem-1515 108 8 trained train VERB fbem-1515 108 9 2 2 NUM fbem-1515 108 10 models model NOUN fbem-1515 108 11 with with ADP fbem-1515 108 12 2 2 NUM fbem-1515 108 13 datasets dataset NOUN fbem-1515 108 14 . . PUNCT fbem-1515 109 1 the the DET fbem-1515 109 2 model model NOUN fbem-1515 109 3 performance performance NOUN fbem-1515 109 4 is be AUX fbem-1515 109 5 shown show VERB fbem-1515 109 6 below below ADP fbem-1515 109 7 : : PUNCT fbem-1515 109 8 with with ADP fbem-1515 109 9 train_1_cleaned train_1_cleaned NUM fbem-1515 109 10 : : PUNCT fbem-1515 109 11 with with ADP fbem-1515 109 12 train_cleaned train_cleane VERB fbem-1515 109 13 : : PUNCT fbem-1515 109 14 the the DET fbem-1515 109 15 auc auc NOUN fbem-1515 109 16 , , PUNCT fbem-1515 109 17 confusion confusion NOUN fbem-1515 109 18 matrix matrix NOUN fbem-1515 109 19 & & CCONJ fbem-1515 109 20 accuracy accuracy NOUN fbem-1515 109 21 of of ADP fbem-1515 109 22 knn knn PROPN fbem-1515 109 23 model model NOUN fbem-1515 109 24 of of ADP fbem-1515 109 25 train_1_cleaned train_1_cleane VERB fbem-1515 109 26 and and CCONJ fbem-1515 109 27 train_cleaned train_cleane VERB fbem-1515 109 28 however however ADV fbem-1515 109 29 , , PUNCT fbem-1515 109 30 considering consider VERB fbem-1515 109 31 the the DET fbem-1515 109 32 model model NOUN fbem-1515 109 33 nature nature NOUN fbem-1515 109 34 of of ADP fbem-1515 109 35 knn knn PROPN fbem-1515 109 36 , , PUNCT fbem-1515 109 37 we we PRON fbem-1515 109 38 decided decide VERB fbem-1515 109 39 to to PART fbem-1515 109 40 fit fit VERB fbem-1515 109 41 our our PRON fbem-1515 109 42 final final ADJ fbem-1515 109 43 model model NOUN fbem-1515 109 44 with with ADP fbem-1515 109 45 β€˜ ' PUNCT fbem-1515 109 46 train train NOUN fbem-1515 109 47 ’ ' PUNCT fbem-1515 109 48 ( ( PUNCT fbem-1515 109 49 k k NOUN fbem-1515 109 50 = = NOUN fbem-1515 109 51 71 71 NUM fbem-1515 109 52 and and CCONJ fbem-1515 109 53 p p X fbem-1515 109 54 = = NOUN fbem-1515 109 55 1 1 NUM fbem-1515 109 56 ) ) PUNCT fbem-1515 109 57 which which PRON fbem-1515 109 58 shares share VERB fbem-1515 109 59 the the DET fbem-1515 109 60 similar similar ADJ fbem-1515 109 61 label label NOUN fbem-1515 109 62 compositions composition NOUN fbem-1515 109 63 as as ADP fbem-1515 109 64 the the DET fbem-1515 109 65 real real ADJ fbem-1515 109 66 business business NOUN fbem-1515 109 67 data datum NOUN fbem-1515 109 68 . . PUNCT fbem-1515 110 1 the the DET fbem-1515 110 2 model model NOUN fbem-1515 110 3 performance performance NOUN fbem-1515 110 4 on on ADP fbem-1515 110 5 validation validation NOUN fbem-1515 110 6 set set VERB fbem-1515 110 7 by by ADP fbem-1515 110 8 default default NOUN fbem-1515 110 9 threshold threshold NOUN fbem-1515 110 10 is be AUX fbem-1515 110 11 ( ( PUNCT fbem-1515 110 12 note note VERB fbem-1515 110 13 that that SCONJ fbem-1515 110 14 positive positive ADJ fbem-1515 110 15 :1 :1 NOUN fbem-1515 110 16 , , PUNCT fbem-1515 110 17 negative:0 negative:0 NOUN fbem-1515 110 18 ): ): PUNCT fbem-1515 110 19 ( ( PUNCT fbem-1515 110 20 fit fit ADJ fbem-1515 110 21 with with ADP fbem-1515 110 22 train_cleaned train_cleane VERB fbem-1515 110 23 ) ) PUNCT fbem-1515 110 24 the the DET fbem-1515 110 25 auc auc NOUN fbem-1515 110 26 , , PUNCT fbem-1515 110 27 confusion confusion NOUN fbem-1515 110 28 matrix matrix NOUN fbem-1515 110 29 & & CCONJ fbem-1515 110 30 accuracy accuracy NOUN fbem-1515 110 31 of of ADP fbem-1515 110 32 knn knn PROPN fbem-1515 110 33 model model NOUN fbem-1515 110 34 of of ADP fbem-1515 110 35 train_cleaned train_cleane VERB fbem-1515 110 36 147 147 NUM fbem-1515 110 37 after after ADP fbem-1515 110 38 getting get VERB fbem-1515 110 39 the the DET fbem-1515 110 40 model model NOUN fbem-1515 110 41 , , PUNCT fbem-1515 110 42 we we PRON fbem-1515 110 43 adopt adopt VERB fbem-1515 110 44 cost cost NOUN fbem-1515 110 45 and and CCONJ fbem-1515 110 46 benefit benefit VERB fbem-1515 110 47 analysis analysis NOUN fbem-1515 110 48 on on ADP fbem-1515 110 49 the the DET fbem-1515 110 50 validation validation NOUN fbem-1515 110 51 set set NOUN fbem-1515 110 52 . . PUNCT fbem-1515 111 1 the the DET fbem-1515 111 2 optimal optimal ADJ fbem-1515 111 3 threshold threshold NOUN fbem-1515 111 4 ( ( PUNCT fbem-1515 111 5 for for ADP fbem-1515 111 6 β€˜ ' PUNCT fbem-1515 111 7 0 0 NUM fbem-1515 111 8 ’ ' PUNCT fbem-1515 111 9 prediction prediction NOUN fbem-1515 111 10 ) ) PUNCT fbem-1515 111 11 is be AUX fbem-1515 111 12 found find VERB fbem-1515 111 13 to to PART fbem-1515 111 14 be be AUX fbem-1515 111 15 0.9 0.9 NUM fbem-1515 111 16 . . PUNCT fbem-1515 112 1 the the DET fbem-1515 112 2 β€˜ ' PUNCT fbem-1515 112 3 cost cost NOUN fbem-1515 112 4 curve curve NOUN fbem-1515 112 5 ’ ' PUNCT fbem-1515 112 6 and and CCONJ fbem-1515 112 7 β€˜ ' PUNCT fbem-1515 112 8 cost cost NOUN fbem-1515 112 9 comparison comparison NOUN fbem-1515 112 10 ’ ' PUNCT fbem-1515 112 11 are be AUX fbem-1515 112 12 shown show VERB fbem-1515 112 13 below below ADP fbem-1515 112 14 : : PUNCT fbem-1515 112 15 cost cost NOUN fbem-1515 112 16 graph graph NOUN fbem-1515 112 17 with with ADP fbem-1515 112 18 different different ADJ fbem-1515 112 19 threshold threshold NOUN fbem-1515 112 20 & & CCONJ fbem-1515 112 21 cost cost VERB fbem-1515 112 22 comparison comparison NOUN fbem-1515 112 23 table table NOUN fbem-1515 112 24 between between ADP fbem-1515 112 25 all all PRON fbem-1515 112 26 predicted predict VERB fbem-1515 112 27 by by ADP fbem-1515 112 28 majority majority NOUN fbem-1515 112 29 , , PUNCT fbem-1515 112 30 predicted predict VERB fbem-1515 112 31 without without ADP fbem-1515 112 32 threshold threshold NOUN fbem-1515 112 33 adjustment adjustment NOUN fbem-1515 112 34 , , PUNCT fbem-1515 112 35 and and CCONJ fbem-1515 112 36 by by ADP fbem-1515 112 37 our our PRON fbem-1515 112 38 0.9 0.9 NUM fbem-1515 112 39 ideal ideal NOUN fbem-1515 112 40 threshold threshold VERB fbem-1515 112 41 the the DET fbem-1515 112 42 cost cost NOUN fbem-1515 112 43 incurred incur VERB fbem-1515 112 44 by by ADP fbem-1515 112 45 a a DET fbem-1515 112 46 majority majority NOUN fbem-1515 112 47 classifier classifier NOUN fbem-1515 112 48 ( ( PUNCT fbem-1515 112 49 all all PRON fbem-1515 112 50 predict predict VERB fbem-1515 112 51 to to PART fbem-1515 112 52 be be AUX fbem-1515 112 53 non non ADJ fbem-1515 112 54 - - NOUN fbem-1515 112 55 default default NOUN fbem-1515 112 56 ) ) PUNCT fbem-1515 112 57 is be AUX fbem-1515 112 58 about about ADV fbem-1515 112 59 43.9 43.9 NUM fbem-1515 112 60 million million NUM fbem-1515 112 61 ( ( PUNCT fbem-1515 112 62 β€œ " PUNCT fbem-1515 112 63 total total ADJ fbem-1515 112 64 cost cost NOUN fbem-1515 112 65 ” " PUNCT fbem-1515 112 66 ) ) PUNCT fbem-1515 112 67 . . PUNCT fbem-1515 113 1 the the DET fbem-1515 113 2 cost cost NOUN fbem-1515 113 3 decreased decrease VERB fbem-1515 113 4 by by ADP fbem-1515 113 5 6.09 6.09 NUM fbem-1515 113 6 % % NOUN fbem-1515 113 7 after after ADP fbem-1515 113 8 the the DET fbem-1515 113 9 prediction prediction NOUN fbem-1515 113 10 with with ADP fbem-1515 113 11 default default NOUN fbem-1515 113 12 threshold threshold NOUN fbem-1515 113 13 , , PUNCT fbem-1515 113 14 and and CCONJ fbem-1515 113 15 further far ADV fbem-1515 113 16 decreased decrease VERB fbem-1515 113 17 by by ADP fbem-1515 113 18 14.81 14.81 NUM fbem-1515 113 19 % % NOUN fbem-1515 113 20 with with ADP fbem-1515 113 21 0.9 0.9 NUM fbem-1515 113 22 threshold threshold NOUN fbem-1515 113 23 . . PUNCT fbem-1515 114 1 evaluation evaluation NOUN fbem-1515 114 2 we we PRON fbem-1515 114 3 have have AUX fbem-1515 114 4 applied apply VERB fbem-1515 114 5 the the DET fbem-1515 114 6 model model NOUN fbem-1515 114 7 and and CCONJ fbem-1515 114 8 the the DET fbem-1515 114 9 optimal optimal ADJ fbem-1515 114 10 threshold threshold NOUN fbem-1515 114 11 on on ADP fbem-1515 114 12 test test NOUN fbem-1515 114 13 set set VERB fbem-1515 114 14 and and CCONJ fbem-1515 114 15 get get VERB fbem-1515 114 16 the the DET fbem-1515 114 17 results result NOUN fbem-1515 114 18 as as SCONJ fbem-1515 114 19 follows follow VERB fbem-1515 114 20 : : PUNCT fbem-1515 114 21 the the DET fbem-1515 114 22 auc auc NOUN fbem-1515 114 23 & & CCONJ fbem-1515 114 24 reduced reduce VERB fbem-1515 114 25 cost cost NOUN fbem-1515 114 26 result result NOUN fbem-1515 114 27 of of ADP fbem-1515 114 28 test test NOUN fbem-1515 114 29 set set VERB fbem-1515 114 30 based base VERB fbem-1515 114 31 on on ADP fbem-1515 114 32 knn knn PROPN fbem-1515 114 33 model model PROPN fbem-1515 114 34 the the DET fbem-1515 114 35 cost cost NOUN fbem-1515 114 36 is be AUX fbem-1515 114 37 reduced reduce VERB fbem-1515 114 38 by by ADP fbem-1515 114 39 about about ADV fbem-1515 114 40 23.26 23.26 NUM fbem-1515 114 41 % % NOUN fbem-1515 114 42 after after ADP fbem-1515 114 43 applying apply VERB fbem-1515 114 44 the the DET fbem-1515 114 45 model model NOUN fbem-1515 114 46 . . PUNCT fbem-1515 115 1 knn knn PROPN fbem-1515 115 2 model model PROPN fbem-1515 115 3 can can AUX fbem-1515 115 4 fit fit VERB fbem-1515 115 5 the the DET fbem-1515 115 6 data datum NOUN fbem-1515 115 7 well well INTJ fbem-1515 115 8 while while SCONJ fbem-1515 115 9 also also ADV fbem-1515 115 10 have have VERB fbem-1515 115 11 limitations limitation NOUN fbem-1515 115 12 . . PUNCT fbem-1515 116 1 ( ( PUNCT fbem-1515 116 2 e.g. e.g. ADV fbem-1515 116 3 the the DET fbem-1515 116 4 actual actual ADJ fbem-1515 116 5 business business NOUN fbem-1515 116 6 data datum NOUN fbem-1515 116 7 volume volume NOUN fbem-1515 116 8 and and CCONJ fbem-1515 116 9 feature feature NOUN fbem-1515 116 10 size size NOUN fbem-1515 116 11 are be AUX fbem-1515 116 12 large large ADJ fbem-1515 116 13 ) ) PUNCT fbem-1515 116 14 4.5 4.5 NUM fbem-1515 116 15 . . PUNCT fbem-1515 117 1 summary summary NOUN fbem-1515 117 2 of of ADP fbem-1515 117 3 performance performance NOUN fbem-1515 117 4 we we PRON fbem-1515 117 5 ’ve ’ve AUX fbem-1515 117 6 tried try VERB fbem-1515 117 7 3 3 NUM fbem-1515 117 8 types type NOUN fbem-1515 117 9 of of ADP fbem-1515 117 10 different different ADJ fbem-1515 117 11 models model NOUN fbem-1515 117 12 , , PUNCT fbem-1515 117 13 namely namely ADV fbem-1515 117 14 , , PUNCT fbem-1515 117 15 decision decision NOUN fbem-1515 117 16 tree tree NOUN fbem-1515 117 17 model model NOUN fbem-1515 117 18 , , PUNCT fbem-1515 117 19 logistic logistic ADJ fbem-1515 117 20 model model NOUN fbem-1515 117 21 and and CCONJ fbem-1515 117 22 knn knn PROPN fbem-1515 117 23 model model PROPN fbem-1515 117 24 , , PUNCT fbem-1515 117 25 and and CCONJ fbem-1515 117 26 we we PRON fbem-1515 117 27 got get VERB fbem-1515 117 28 3 3 NUM fbem-1515 117 29 different different ADJ fbem-1515 117 30 evaluations evaluation NOUN fbem-1515 117 31 based base VERB fbem-1515 117 32 on on ADP fbem-1515 117 33 them they PRON fbem-1515 117 34 . . PUNCT fbem-1515 118 1 firstly firstly ADV fbem-1515 118 2 , , PUNCT fbem-1515 118 3 the the DET fbem-1515 118 4 decision decision NOUN fbem-1515 118 5 tree tree NOUN fbem-1515 118 6 model model NOUN fbem-1515 118 7 , , PUNCT fbem-1515 118 8 after after ADP fbem-1515 118 9 doing do VERB fbem-1515 118 10 the the DET fbem-1515 118 11 hyperparameter hyperparameter NOUN fbem-1515 118 12 tuning tuning NOUN fbem-1515 118 13 , , PUNCT fbem-1515 118 14 the the DET fbem-1515 118 15 auc auc NOUN fbem-1515 118 16 of of ADP fbem-1515 118 17 our our PRON fbem-1515 118 18 finalized finalized ADJ fbem-1515 118 19 model model NOUN fbem-1515 118 20 is be AUX fbem-1515 118 21 0.525804 0.525804 PROPN fbem-1515 118 22 based base VERB fbem-1515 118 23 on on ADP fbem-1515 118 24 our our PRON fbem-1515 118 25 separate separate ADJ fbem-1515 118 26 test test NOUN fbem-1515 118 27 set set VERB fbem-1515 118 28 . . PUNCT fbem-1515 119 1 after after ADP fbem-1515 119 2 doing do VERB fbem-1515 119 3 the the DET fbem-1515 119 4 cost cost NOUN fbem-1515 119 5 analysis analysis NOUN fbem-1515 119 6 , , PUNCT fbem-1515 119 7 we we PRON fbem-1515 119 8 used use VERB fbem-1515 119 9 the the DET fbem-1515 119 10 threshold threshold NOUN fbem-1515 119 11 of of ADP fbem-1515 119 12 0.93 0.93 NUM fbem-1515 119 13 and and CCONJ fbem-1515 119 14 under under ADP fbem-1515 119 15 this this DET fbem-1515 119 16 threshold threshold NOUN fbem-1515 119 17 our our PRON fbem-1515 119 18 cost cost NOUN fbem-1515 119 19 can can AUX fbem-1515 119 20 be be AUX fbem-1515 119 21 reduced reduce VERB fbem-1515 119 22 by by ADP fbem-1515 119 23 22.98 22.98 NUM fbem-1515 119 24 % % NOUN fbem-1515 119 25 in in ADP fbem-1515 119 26 our our PRON fbem-1515 119 27 separate separate ADJ fbem-1515 119 28 test test NOUN fbem-1515 119 29 set set NOUN fbem-1515 119 30 compared compare VERB fbem-1515 119 31 to to ADP fbem-1515 119 32 majority majority NOUN fbem-1515 119 33 classifier classifier NOUN fbem-1515 119 34 ( ( PUNCT fbem-1515 119 35 all all PRON fbem-1515 119 36 predict predict VERB fbem-1515 119 37 to to PART fbem-1515 119 38 be be AUX fbem-1515 119 39 non non ADJ fbem-1515 119 40 - - NOUN fbem-1515 119 41 default default ADJ fbem-1515 119 42 ) ) PUNCT fbem-1515 119 43 . . PUNCT fbem-1515 120 1 secondly secondly ADV fbem-1515 120 2 , , PUNCT fbem-1515 120 3 the the DET fbem-1515 120 4 logistic logistic ADJ fbem-1515 120 5 model model NOUN fbem-1515 120 6 , , PUNCT fbem-1515 120 7 after after ADP fbem-1515 120 8 doing do VERB fbem-1515 120 9 the the DET fbem-1515 120 10 hyperparameter hyperparameter NOUN fbem-1515 120 11 tuning tuning NOUN fbem-1515 120 12 , , PUNCT fbem-1515 120 13 the the DET fbem-1515 120 14 auc auc NOUN fbem-1515 120 15 of of ADP fbem-1515 120 16 our our PRON fbem-1515 120 17 finalized finalized ADJ fbem-1515 120 18 model model NOUN fbem-1515 120 19 is be AUX fbem-1515 120 20 0.500804 0.500804 NUM fbem-1515 120 21 based base VERB fbem-1515 120 22 on on ADP fbem-1515 120 23 our our PRON fbem-1515 120 24 separate separate ADJ fbem-1515 120 25 test test NOUN fbem-1515 120 26 set set VERB fbem-1515 120 27 . . PUNCT fbem-1515 121 1 after after ADP fbem-1515 121 2 doing do VERB fbem-1515 121 3 the the DET fbem-1515 121 4 cost cost NOUN fbem-1515 121 5 analysis analysis NOUN fbem-1515 121 6 , , PUNCT fbem-1515 121 7 we we PRON fbem-1515 121 8 used use VERB fbem-1515 121 9 the the DET fbem-1515 121 10 threshold threshold NOUN fbem-1515 121 11 of of ADP fbem-1515 121 12 0.9 0.9 NUM fbem-1515 121 13 and and CCONJ fbem-1515 121 14 under under ADP fbem-1515 121 15 this this DET fbem-1515 121 16 threshold threshold NOUN fbem-1515 121 17 our our PRON fbem-1515 121 18 cost cost NOUN fbem-1515 121 19 can can AUX fbem-1515 121 20 be be AUX fbem-1515 121 21 reduced reduce VERB fbem-1515 121 22 by by ADP fbem-1515 121 23 26.6 26.6 NUM fbem-1515 121 24 % % NOUN fbem-1515 121 25 in in ADP fbem-1515 121 26 our our PRON fbem-1515 121 27 separate separate ADJ fbem-1515 121 28 test test NOUN fbem-1515 121 29 set set NOUN fbem-1515 121 30 compared compare VERB fbem-1515 121 31 to to ADP fbem-1515 121 32 majority majority NOUN fbem-1515 121 33 classifier classifier NOUN fbem-1515 121 34 ( ( PUNCT fbem-1515 121 35 all all PRON fbem-1515 121 36 predict predict VERB fbem-1515 121 37 to to PART fbem-1515 121 38 be be AUX fbem-1515 121 39 non non ADJ fbem-1515 121 40 - - NOUN fbem-1515 121 41 default default NOUN fbem-1515 121 42 ) ) PUNCT fbem-1515 121 43 . . PUNCT fbem-1515 122 1 lastly lastly ADV fbem-1515 122 2 , , PUNCT fbem-1515 122 3 the the DET fbem-1515 122 4 knn knn PROPN fbem-1515 122 5 model model PROPN fbem-1515 122 6 , , PUNCT fbem-1515 122 7 after after ADP fbem-1515 122 8 doing do VERB fbem-1515 122 9 the the DET fbem-1515 122 10 tuning tuning NOUN fbem-1515 122 11 for for ADP fbem-1515 122 12 the the DET fbem-1515 122 13 values value NOUN fbem-1515 122 14 of of ADP fbem-1515 122 15 k k PROPN fbem-1515 122 16 and and CCONJ fbem-1515 122 17 p p X fbem-1515 122 18 , , PUNCT fbem-1515 122 19 the the DET fbem-1515 122 20 auc auc NOUN fbem-1515 122 21 of of ADP fbem-1515 122 22 our our PRON fbem-1515 122 23 finalized finalized ADJ fbem-1515 122 24 model model NOUN fbem-1515 122 25 is be AUX fbem-1515 122 26 0.704615 0.704615 NUM fbem-1515 122 27 based base VERB fbem-1515 122 28 on on ADP fbem-1515 122 29 our our PRON fbem-1515 122 30 separate separate ADJ fbem-1515 122 31 test test NOUN fbem-1515 122 32 set set VERB fbem-1515 122 33 . . PUNCT fbem-1515 123 1 after after ADP fbem-1515 123 2 doing do VERB fbem-1515 123 3 the the DET fbem-1515 123 4 cost cost NOUN fbem-1515 123 5 analysis analysis NOUN fbem-1515 123 6 , , PUNCT fbem-1515 123 7 we we PRON fbem-1515 123 8 used use VERB fbem-1515 123 9 the the DET fbem-1515 123 10 threshold threshold NOUN fbem-1515 123 11 of of ADP fbem-1515 123 12 0.9 0.9 NUM fbem-1515 123 13 and and CCONJ fbem-1515 123 14 under under ADP fbem-1515 123 15 this this DET fbem-1515 123 16 threshold threshold NOUN fbem-1515 123 17 our our PRON fbem-1515 123 18 cost cost NOUN fbem-1515 123 19 can can AUX fbem-1515 123 20 be be AUX fbem-1515 123 21 reduced reduce VERB fbem-1515 123 22 by by ADP fbem-1515 123 23 23.26 23.26 NUM fbem-1515 123 24 % % NOUN fbem-1515 123 25 in in ADP fbem-1515 123 26 our our PRON fbem-1515 123 27 separate separate ADJ fbem-1515 123 28 test test NOUN fbem-1515 123 29 set set NOUN fbem-1515 123 30 compared compare VERB fbem-1515 123 31 to to ADP fbem-1515 123 32 majority majority NOUN fbem-1515 123 33 classifier classifier NOUN fbem-1515 123 34 ( ( PUNCT fbem-1515 123 35 all all PRON fbem-1515 123 36 predict predict VERB fbem-1515 123 37 to to PART fbem-1515 123 38 be be AUX fbem-1515 123 39 non non ADJ fbem-1515 123 40 - - NOUN fbem-1515 123 41 default default ADJ fbem-1515 123 42 ) ) PUNCT fbem-1515 123 43 . . PUNCT fbem-1515 124 1 in in ADP fbem-1515 124 2 conclusion conclusion NOUN fbem-1515 124 3 , , PUNCT fbem-1515 124 4 we we PRON fbem-1515 124 5 can can AUX fbem-1515 124 6 find find VERB fbem-1515 124 7 that that SCONJ fbem-1515 124 8 knn knn PROPN fbem-1515 124 9 did do VERB fbem-1515 124 10 the the DET fbem-1515 124 11 best good ADJ fbem-1515 124 12 in in ADP fbem-1515 124 13 predicting predict VERB fbem-1515 124 14 default default NOUN fbem-1515 124 15 situations situation NOUN fbem-1515 124 16 under under ADP fbem-1515 124 17 all all DET fbem-1515 124 18 thresholds threshold NOUN fbem-1515 124 19 overall overall ADJ fbem-1515 124 20 with with ADP fbem-1515 124 21 the the DET fbem-1515 124 22 biggest big ADJ fbem-1515 124 23 auc auc NOUN fbem-1515 124 24 of of ADP fbem-1515 124 25 0.704615 0.704615 NUM fbem-1515 124 26 on on ADP fbem-1515 124 27 test test NOUN fbem-1515 124 28 set set VERB fbem-1515 124 29 . . PUNCT fbem-1515 125 1 but but CCONJ fbem-1515 125 2 it it PRON fbem-1515 125 3 did do AUX fbem-1515 125 4 n’t not PART fbem-1515 125 5 do do VERB fbem-1515 125 6 that that PRON fbem-1515 125 7 well well ADV fbem-1515 125 8 in in ADP fbem-1515 125 9 cost cost NOUN fbem-1515 125 10 analysis analysis NOUN fbem-1515 125 11 under under ADP fbem-1515 125 12 the the DET fbem-1515 125 13 specific specific ADJ fbem-1515 125 14 threshold threshold NOUN fbem-1515 125 15 of of ADP fbem-1515 125 16 0.9 0.9 NUM fbem-1515 125 17 , , PUNCT fbem-1515 125 18 logistic logistic ADJ fbem-1515 125 19 model model NOUN fbem-1515 125 20 did do VERB fbem-1515 125 21 best well ADV fbem-1515 125 22 in in ADP fbem-1515 125 23 reducing reduce VERB fbem-1515 125 24 cost cost NOUN fbem-1515 125 25 with with ADP fbem-1515 125 26 ideal ideal ADJ fbem-1515 125 27 threshold threshold NOUN fbem-1515 125 28 , , PUNCT fbem-1515 125 29 which which PRON fbem-1515 125 30 lead lead VERB fbem-1515 125 31 to to ADP fbem-1515 125 32 the the DET fbem-1515 125 33 biggest big ADJ fbem-1515 125 34 decrease decrease NOUN fbem-1515 125 35 among among ADP fbem-1515 125 36 all all DET fbem-1515 125 37 3 3 NUM fbem-1515 125 38 models model NOUN fbem-1515 125 39 of of ADP fbem-1515 125 40 26.6 26.6 NUM fbem-1515 125 41 % % NOUN fbem-1515 125 42 , , PUNCT fbem-1515 125 43 but but CCONJ fbem-1515 125 44 the the DET fbem-1515 125 45 auc auc NOUN fbem-1515 125 46 of of ADP fbem-1515 125 47 it it PRON fbem-1515 125 48 is be AUX fbem-1515 125 49 not not PART fbem-1515 125 50 as as ADV fbem-1515 125 51 high high ADJ fbem-1515 125 52 . . PUNCT fbem-1515 126 1 so so ADV fbem-1515 126 2 , , PUNCT fbem-1515 126 3 we we PRON fbem-1515 126 4 can can AUX fbem-1515 126 5 further far ADV fbem-1515 126 6 ensemble ensemble VERB fbem-1515 126 7 them they PRON fbem-1515 126 8 together together ADV fbem-1515 126 9 to to PART fbem-1515 126 10 see see VERB fbem-1515 126 11 whether whether SCONJ fbem-1515 126 12 we we PRON fbem-1515 126 13 can can AUX fbem-1515 126 14 achieve achieve VERB fbem-1515 126 15 an an DET fbem-1515 126 16 even even ADV fbem-1515 126 17 better well ADJ fbem-1515 126 18 model model NOUN fbem-1515 126 19 doing do VERB fbem-1515 126 20 well well ADV fbem-1515 126 21 both both CCONJ fbem-1515 126 22 in in ADP fbem-1515 126 23 prediction prediction NOUN fbem-1515 126 24 and and CCONJ fbem-1515 126 25 reducing reduce VERB fbem-1515 126 26 cost cost NOUN fbem-1515 126 27 . . PUNCT fbem-1515 127 1 5 5 X fbem-1515 127 2 . . X fbem-1515 127 3 ensemble ensemble ADJ fbem-1515 127 4 model model NOUN fbem-1515 127 5 we we PRON fbem-1515 127 6 have have AUX fbem-1515 127 7 incorporated incorporate VERB fbem-1515 127 8 the the DET fbem-1515 127 9 3 3 NUM fbem-1515 127 10 models model NOUN fbem-1515 127 11 elaborated elaborate VERB fbem-1515 127 12 above above ADV fbem-1515 127 13 and and CCONJ fbem-1515 127 14 tried try VERB fbem-1515 127 15 2 2 NUM fbem-1515 127 16 other other ADJ fbem-1515 127 17 new new ADJ fbem-1515 127 18 models model NOUN fbem-1515 127 19 ( ( PUNCT fbem-1515 127 20 random random ADJ fbem-1515 127 21 forest forest NOUN fbem-1515 127 22 and and CCONJ fbem-1515 127 23 adaptive adaptive ADJ fbem-1515 127 24 boost boost NOUN fbem-1515 127 25 ) ) PUNCT fbem-1515 127 26 to to PART fbem-1515 127 27 collectively collectively ADV fbem-1515 127 28 predict predict VERB fbem-1515 127 29 β€˜ ' PUNCT fbem-1515 127 30 default default NOUN fbem-1515 127 31 ’ ' PUNCT fbem-1515 127 32 . . PUNCT fbem-1515 128 1 results result NOUN fbem-1515 128 2 show show VERB fbem-1515 128 3 that that SCONJ fbem-1515 128 4 random random ADJ fbem-1515 128 5 forest forest NOUN fbem-1515 128 6 works work VERB fbem-1515 128 7 well well ADV fbem-1515 128 8 on on ADP fbem-1515 128 9 the the DET fbem-1515 128 10 data datum NOUN fbem-1515 128 11 while while SCONJ fbem-1515 128 12 adaptive adaptive ADJ fbem-1515 128 13 boost boost NOUN fbem-1515 128 14 does do VERB fbem-1515 128 15 not not PART fbem-1515 128 16 . . PUNCT fbem-1515 129 1 thus thus ADV fbem-1515 129 2 , , PUNCT fbem-1515 129 3 we we PRON fbem-1515 129 4 discard discard VERB fbem-1515 129 5 adaptive adaptive ADJ fbem-1515 129 6 boost boost NOUN fbem-1515 129 7 model model NOUN fbem-1515 129 8 and and CCONJ fbem-1515 129 9 use use VERB fbem-1515 129 10 majority majority NOUN fbem-1515 129 11 voting voting NOUN fbem-1515 129 12 classifier classifier NOUN fbem-1515 129 13 to to PART fbem-1515 129 14 build build VERB fbem-1515 129 15 the the DET fbem-1515 129 16 ensemble ensemble ADJ fbem-1515 129 17 model model NOUN fbem-1515 129 18 in in ADP fbem-1515 129 19 2 2 NUM fbem-1515 129 20 versions version NOUN fbem-1515 129 21 . . PUNCT fbem-1515 130 1 next next ADV fbem-1515 130 2 , , PUNCT fbem-1515 130 3 we we PRON fbem-1515 130 4 have have AUX fbem-1515 130 5 also also ADV fbem-1515 130 6 explored explore VERB fbem-1515 130 7 how how SCONJ fbem-1515 130 8 it it PRON fbem-1515 130 9 performs perform VERB fbem-1515 130 10 with with ADP fbem-1515 130 11 and and CCONJ fbem-1515 130 12 without without ADP fbem-1515 130 13 the the DET fbem-1515 130 14 prediction prediction NOUN fbem-1515 130 15 threshold threshold NOUN fbem-1515 130 16 ( ( PUNCT fbem-1515 130 17 0.9 0.9 NUM fbem-1515 130 18 ) ) PUNCT fbem-1515 130 19 on on ADP fbem-1515 130 20 the the DET fbem-1515 130 21 test test NOUN fbem-1515 130 22 set set NOUN fbem-1515 130 23 . . PUNCT fbem-1515 131 1 the the DET fbem-1515 131 2 test test NOUN fbem-1515 131 3 results result VERB fbem-1515 131 4 for for ADP fbem-1515 131 5 prediction prediction NOUN fbem-1515 131 6 with with ADP fbem-1515 131 7 threshold threshold NOUN fbem-1515 131 8 0.9 0.9 NUM fbem-1515 131 9 ( ( PUNCT fbem-1515 131 10 shown show VERB fbem-1515 131 11 below below ADV fbem-1515 131 12 ) ) PUNCT fbem-1515 131 13 are be AUX fbem-1515 131 14 much much ADV fbem-1515 131 15 better well ADJ fbem-1515 131 16 than than ADP fbem-1515 131 17 the the DET fbem-1515 131 18 default default NOUN fbem-1515 131 19 prediction prediction NOUN fbem-1515 131 20 . . PUNCT fbem-1515 132 1 5.1 5.1 NUM fbem-1515 132 2 . . PUNCT fbem-1515 132 3 version version NOUN fbem-1515 132 4 1 1 NUM fbem-1515 132 5 : : PUNCT fbem-1515 132 6 knn knn PROPN fbem-1515 132 7 , , PUNCT fbem-1515 132 8 logistic logistic ADJ fbem-1515 132 9 , , PUNCT fbem-1515 132 10 decision decision NOUN fbem-1515 132 11 tree tree NOUN fbem-1515 132 12 , , PUNCT fbem-1515 132 13 random random ADJ fbem-1515 132 14 forest forest NOUN fbem-1515 132 15 prediction prediction NOUN fbem-1515 132 16 1 1 NUM fbem-1515 132 17 ( ( PUNCT fbem-1515 132 18 majority majority NOUN fbem-1515 132 19 prediction prediction NOUN fbem-1515 132 20 ) ) PUNCT fbem-1515 132 21 prediction prediction NOUN fbem-1515 132 22 2 2 NUM fbem-1515 132 23 ( ( PUNCT fbem-1515 132 24 β€˜ ' PUNCT fbem-1515 132 25 1 1 NUM fbem-1515 132 26 ’ ' PUNCT fbem-1515 132 27 if if SCONJ fbem-1515 132 28 > > X fbem-1515 132 29 = = SYM fbem-1515 132 30 0.5 0.5 NUM fbem-1515 132 31 models model NOUN fbem-1515 132 32 predict predict VERB fbem-1515 132 33 β€˜ ' PUNCT fbem-1515 132 34 1 1 NUM fbem-1515 132 35 ’ ' PUNCT fbem-1515 132 36 ) ) PUNCT fbem-1515 132 37 cost cost VERB fbem-1515 132 38 comparison comparison NOUN fbem-1515 132 39 table table NOUN fbem-1515 132 40 between between ADP fbem-1515 132 41 2 2 NUM fbem-1515 132 42 prediction prediction NOUN fbem-1515 132 43 methods method NOUN fbem-1515 132 44 by by ADP fbem-1515 132 45 our our PRON fbem-1515 132 46 0.9 0.9 NUM fbem-1515 132 47 ideal ideal ADJ fbem-1515 132 48 threshold threshold NOUN fbem-1515 132 49 for for ADP fbem-1515 132 50 fusion fusion NOUN fbem-1515 132 51 model model NOUN fbem-1515 132 52 version version NOUN fbem-1515 132 53 1 1 NUM fbem-1515 132 54 we we PRON fbem-1515 132 55 have have AUX fbem-1515 132 56 included include VERB fbem-1515 132 57 all all DET fbem-1515 132 58 models model NOUN fbem-1515 132 59 at at ADP fbem-1515 132 60 first first ADV fbem-1515 132 61 and and CCONJ fbem-1515 132 62 tried try VERB fbem-1515 132 63 two two NUM fbem-1515 132 64 types type NOUN fbem-1515 132 65 of of ADP fbem-1515 132 66 prediction prediction NOUN fbem-1515 132 67 ( ( PUNCT fbem-1515 132 68 > > X fbem-1515 132 69 or or CCONJ fbem-1515 132 70 > > X fbem-1515 132 71 = = NUM fbem-1515 132 72 0.5 0.5 NUM fbem-1515 132 73 ) ) PUNCT fbem-1515 132 74 . . PUNCT fbem-1515 133 1 results result NOUN fbem-1515 133 2 showed show VERB fbem-1515 133 3 that that DET fbem-1515 133 4 prediction prediction NOUN fbem-1515 133 5 2 2 NUM fbem-1515 133 6 works work VERB fbem-1515 133 7 better well ADV fbem-1515 133 8 and and CCONJ fbem-1515 133 9 can can AUX fbem-1515 133 10 reduce reduce VERB fbem-1515 133 11 costs cost NOUN fbem-1515 133 12 by by ADP fbem-1515 133 13 35.98 35.98 NUM fbem-1515 133 14 % % NOUN fbem-1515 133 15 . . PUNCT fbem-1515 134 1 5.2 5.2 NUM fbem-1515 134 2 . . PUNCT fbem-1515 134 3 version version NOUN fbem-1515 134 4 2 2 NUM fbem-1515 134 5 : : PUNCT fbem-1515 134 6 knn knn PROPN fbem-1515 134 7 , , PUNCT fbem-1515 134 8 logistic logistic ADJ fbem-1515 134 9 , , PUNCT fbem-1515 134 10 random random ADJ fbem-1515 134 11 forest forest NOUN fbem-1515 134 12 prediction prediction NOUN fbem-1515 134 13 1(majority 1(majority NUM fbem-1515 134 14 prediction prediction NOUN fbem-1515 134 15 ) ) PUNCT fbem-1515 134 16 148 148 NUM fbem-1515 134 17 prediction prediction NOUN fbem-1515 134 18 2 2 NUM fbem-1515 134 19 ( ( PUNCT fbem-1515 134 20 0 0 NUM fbem-1515 134 21 if if SCONJ fbem-1515 134 22 all all DET fbem-1515 134 23 3 3 NUM fbem-1515 134 24 models model NOUN fbem-1515 134 25 predict predict VERB fbem-1515 134 26 0 0 NUM fbem-1515 134 27 ) ) PUNCT fbem-1515 134 28 cost cost VERB fbem-1515 134 29 comparison comparison NOUN fbem-1515 134 30 table table NOUN fbem-1515 134 31 between between ADP fbem-1515 134 32 2 2 NUM fbem-1515 134 33 prediction prediction NOUN fbem-1515 134 34 methods method NOUN fbem-1515 134 35 by by ADP fbem-1515 134 36 our our PRON fbem-1515 134 37 0.9 0.9 NUM fbem-1515 134 38 ideal ideal ADJ fbem-1515 134 39 threshold threshold NOUN fbem-1515 134 40 for for ADP fbem-1515 134 41 fusion fusion NOUN fbem-1515 134 42 model model NOUN fbem-1515 134 43 version version NOUN fbem-1515 134 44 2 2 NUM fbem-1515 134 45 we we PRON fbem-1515 134 46 exclude exclude VERB fbem-1515 134 47 decision decision NOUN fbem-1515 134 48 tree tree NOUN fbem-1515 134 49 in in ADP fbem-1515 134 50 this this DET fbem-1515 134 51 version version NOUN fbem-1515 134 52 since since SCONJ fbem-1515 134 53 it it PRON fbem-1515 134 54 has have VERB fbem-1515 134 55 a a DET fbem-1515 134 56 relatively relatively ADV fbem-1515 134 57 poor poor ADJ fbem-1515 134 58 performance performance NOUN fbem-1515 134 59 separately separately ADV fbem-1515 134 60 . . PUNCT fbem-1515 135 1 as as SCONJ fbem-1515 135 2 we we PRON fbem-1515 135 3 found find VERB fbem-1515 135 4 that that SCONJ fbem-1515 135 5 separate separate ADJ fbem-1515 135 6 models model NOUN fbem-1515 135 7 fail fail VERB fbem-1515 135 8 to to PART fbem-1515 135 9 predict predict VERB fbem-1515 135 10 more more ADJ fbem-1515 135 11 β€˜ ' PUNCT fbem-1515 135 12 default default NOUN fbem-1515 135 13 ’ ' PUNCT fbem-1515 135 14 than than ADP fbem-1515 135 15 β€˜ ' PUNCT fbem-1515 135 16 nondefault nondefault NOUN fbem-1515 135 17 ’ ' PUNCT fbem-1515 135 18 . . PUNCT fbem-1515 136 1 we we PRON fbem-1515 136 2 try try VERB fbem-1515 136 3 to to PART fbem-1515 136 4 recognize recognize VERB fbem-1515 136 5 all all DET fbem-1515 136 6 the the DET fbem-1515 136 7 default default NOUN fbem-1515 136 8 predictions prediction NOUN fbem-1515 136 9 ( ( PUNCT fbem-1515 136 10 pred pre VERB fbem-1515 136 11 2 2 NUM fbem-1515 136 12 ) ) PUNCT fbem-1515 136 13 . . PUNCT fbem-1515 137 1 after after ADP fbem-1515 137 2 comparing compare VERB fbem-1515 137 3 both both PRON fbem-1515 137 4 , , PUNCT fbem-1515 137 5 we we PRON fbem-1515 137 6 found find VERB fbem-1515 137 7 that that SCONJ fbem-1515 137 8 majority majority NOUN fbem-1515 137 9 predictions prediction NOUN fbem-1515 137 10 work work VERB fbem-1515 137 11 better well ADV fbem-1515 137 12 and and CCONJ fbem-1515 137 13 can can AUX fbem-1515 137 14 reduce reduce VERB fbem-1515 137 15 costs cost NOUN fbem-1515 137 16 by by ADP fbem-1515 137 17 38.9 38.9 NUM fbem-1515 137 18 % % NOUN fbem-1515 137 19 . . PUNCT fbem-1515 138 1 5.3 5.3 NUM fbem-1515 138 2 . . PUNCT fbem-1515 139 1 conclusion conclusion NOUN fbem-1515 139 2 ensemble ensemble ADJ fbem-1515 139 3 model model NOUN fbem-1515 139 4 can can AUX fbem-1515 139 5 greatly greatly ADV fbem-1515 139 6 improve improve VERB fbem-1515 139 7 the the DET fbem-1515 139 8 prediction prediction NOUN fbem-1515 139 9 quality quality NOUN fbem-1515 139 10 and and CCONJ fbem-1515 139 11 boost boost VERB fbem-1515 139 12 the the DET fbem-1515 139 13 cost cost NOUN fbem-1515 139 14 reduction reduction NOUN fbem-1515 139 15 compared compare VERB fbem-1515 139 16 with with ADP fbem-1515 139 17 a a DET fbem-1515 139 18 single single ADJ fbem-1515 139 19 model model NOUN fbem-1515 139 20 . . PUNCT fbem-1515 140 1 among among ADP fbem-1515 140 2 all all DET fbem-1515 140 3 the the DET fbem-1515 140 4 combinations combination NOUN fbem-1515 140 5 we we PRON fbem-1515 140 6 ’ve ’ve AUX fbem-1515 140 7 tested test VERB fbem-1515 140 8 , , PUNCT fbem-1515 140 9 ensemble ensemble ADJ fbem-1515 140 10 model model NOUN fbem-1515 140 11 ( ( PUNCT fbem-1515 140 12 knn knn PROPN fbem-1515 140 13 , , PUNCT fbem-1515 140 14 logistic logistic ADJ fbem-1515 140 15 and and CCONJ fbem-1515 140 16 random random ADJ fbem-1515 140 17 forest forest NOUN fbem-1515 140 18 ) ) PUNCT fbem-1515 140 19 with with ADP fbem-1515 140 20 majority majority NOUN fbem-1515 140 21 prediction prediction NOUN fbem-1515 140 22 and and CCONJ fbem-1515 140 23 0.9 0.9 NUM fbem-1515 140 24 threshold threshold NOUN fbem-1515 140 25 performs perform VERB fbem-1515 140 26 the the DET fbem-1515 140 27 best good ADJ fbem-1515 140 28 . . PUNCT fbem-1515 141 1 6 6 X fbem-1515 141 2 . . X fbem-1515 141 3 business business NOUN fbem-1515 141 4 application application NOUN fbem-1515 141 5 data datum NOUN fbem-1515 141 6 collection collection NOUN fbem-1515 141 7 advice advice NOUN fbem-1515 141 8 i. i. PROPN fbem-1515 141 9 logistic logistic PROPN fbem-1515 141 10 : : PUNCT fbem-1515 141 11 we we PRON fbem-1515 141 12 have have VERB fbem-1515 141 13 the the DET fbem-1515 141 14 table table NOUN fbem-1515 141 15 of of ADP fbem-1515 141 16 coefficients coefficient NOUN fbem-1515 141 17 and and CCONJ fbem-1515 141 18 intercept intercept NOUN fbem-1515 141 19 as as SCONJ fbem-1515 141 20 shown show VERB fbem-1515 141 21 : : PUNCT fbem-1515 141 22 table table NOUN fbem-1515 141 23 of of ADP fbem-1515 141 24 coefficient coefficient NOUN fbem-1515 141 25 and and CCONJ fbem-1515 141 26 intercepts intercept NOUN fbem-1515 141 27 of of ADP fbem-1515 141 28 our our PRON fbem-1515 141 29 logistic logistic ADJ fbem-1515 141 30 model model NOUN fbem-1515 141 31 as as ADP fbem-1515 141 32 the the DET fbem-1515 141 33 table table NOUN fbem-1515 141 34 shown show VERB fbem-1515 141 35 , , PUNCT fbem-1515 141 36 we we PRON fbem-1515 141 37 can can AUX fbem-1515 141 38 find find VERB fbem-1515 141 39 that that SCONJ fbem-1515 141 40 the the DET fbem-1515 141 41 features feature NOUN fbem-1515 141 42 with with ADP fbem-1515 141 43 5 5 NUM fbem-1515 141 44 - - PUNCT fbem-1515 141 45 top top NOUN fbem-1515 141 46 highest high ADJ fbem-1515 141 47 absolute absolute ADJ fbem-1515 141 48 value value NOUN fbem-1515 141 49 of of ADP fbem-1515 141 50 coefficient coefficient NOUN fbem-1515 141 51 ( ( PUNCT fbem-1515 141 52 which which PRON fbem-1515 141 53 means mean VERB fbem-1515 141 54 strong strong ADJ fbem-1515 141 55 impact impact NOUN fbem-1515 141 56 of of ADP fbem-1515 141 57 the the DET fbem-1515 141 58 feature feature NOUN fbem-1515 141 59 on on ADP fbem-1515 141 60 the the DET fbem-1515 141 61 probability probability NOUN fbem-1515 141 62 of of ADP fbem-1515 141 63 prediction prediction NOUN fbem-1515 141 64 outcome outcome NOUN fbem-1515 141 65 being be AUX fbem-1515 141 66 1 1 NUM fbem-1515 141 67 , , PUNCT fbem-1515 141 68 which which PRON fbem-1515 141 69 is be AUX fbem-1515 141 70 default default NOUN fbem-1515 141 71 ) ) PUNCT fbem-1515 141 72 are be AUX fbem-1515 141 73 ' ' PUNCT fbem-1515 141 74 mobile_tag'(mobile mobile_tag'(mobile PROPN fbem-1515 141 75 number number NOUN fbem-1515 141 76 provided provide VERB fbem-1515 141 77 by by ADP fbem-1515 141 78 client client NOUN fbem-1515 141 79 , , PUNCT fbem-1515 141 80 1 1 NUM fbem-1515 141 81 means mean VERB fbem-1515 141 82 yes yes INTJ fbem-1515 141 83 and and CCONJ fbem-1515 141 84 0 0 NUM fbem-1515 141 85 means mean VERB fbem-1515 141 86 no no NOUN fbem-1515 141 87 ) ) PUNCT fbem-1515 141 88 , , PUNCT fbem-1515 141 89 ' ' PUNCT fbem-1515 141 90 score_source_2'(the score_source_2'(the DET fbem-1515 141 91 second second ADJ fbem-1515 141 92 trust trust NOUN fbem-1515 141 93 score score NOUN fbem-1515 141 94 from from ADP fbem-1515 141 95 third third ADJ fbem-1515 141 96 party party NOUN fbem-1515 141 97 ) ) PUNCT fbem-1515 141 98 , , PUNCT fbem-1515 141 99 ' ' PUNCT fbem-1515 141 100 score_source_1_z'(z score_source_1_z'(z ADJ fbem-1515 141 101 score score NOUN fbem-1515 141 102 of of ADP fbem-1515 141 103 the the DET fbem-1515 141 104 first first ADJ fbem-1515 141 105 trust trust NOUN fbem-1515 141 106 score score NOUN fbem-1515 141 107 from from ADP fbem-1515 141 108 third third ADJ fbem-1515 141 109 party party NOUN fbem-1515 141 110 ) ) PUNCT fbem-1515 141 111 , , PUNCT fbem-1515 141 112 ' ' PUNCT fbem-1515 141 113 score_source_3_z'(z score_source_3_z'(z NOUN fbem-1515 141 114 score score NOUN fbem-1515 141 115 of of ADP fbem-1515 141 116 the the DET fbem-1515 141 117 third third ADJ fbem-1515 141 118 trust trust NOUN fbem-1515 141 119 score score NOUN fbem-1515 141 120 from from ADP fbem-1515 141 121 third third ADJ fbem-1515 141 122 party party NOUN fbem-1515 141 123 ) ) PUNCT fbem-1515 141 124 , , PUNCT fbem-1515 141 125 ' ' PUNCT fbem-1515 141 126 client_education_junior client_education_junior ADJ fbem-1515 141 127 secondary'(whether secondary'(whether ADJ fbem-1515 141 128 clients client NOUN fbem-1515 141 129 have have AUX fbem-1515 141 130 finished finish VERB fbem-1515 141 131 the the DET fbem-1515 141 132 junior junior ADJ fbem-1515 141 133 secondary secondary ADJ fbem-1515 141 134 education education NOUN fbem-1515 141 135 ) ) PUNCT fbem-1515 141 136 . . PUNCT fbem-1515 142 1 ii ii PROPN fbem-1515 142 2 . . PUNCT fbem-1515 142 3 decision decision NOUN fbem-1515 142 4 tree tree NOUN fbem-1515 142 5 : : PUNCT fbem-1515 142 6 our our PRON fbem-1515 142 7 top top ADJ fbem-1515 142 8 10 10 NUM fbem-1515 142 9 features feature NOUN fbem-1515 142 10 with with ADP fbem-1515 142 11 the the DET fbem-1515 142 12 most most ADJ fbem-1515 142 13 importance importance NOUN fbem-1515 142 14 are be AUX fbem-1515 142 15 shown show VERB fbem-1515 142 16 : : PUNCT fbem-1515 142 17 top top ADJ fbem-1515 142 18 10 10 NUM fbem-1515 142 19 features feature NOUN fbem-1515 142 20 with with ADP fbem-1515 142 21 the the DET fbem-1515 142 22 most most ADJ fbem-1515 142 23 importance importance NOUN fbem-1515 142 24 of of ADP fbem-1515 142 25 our our PRON fbem-1515 142 26 decision decision NOUN fbem-1515 142 27 tree tree NOUN fbem-1515 142 28 model model NOUN fbem-1515 142 29 surprisingly surprisingly ADV fbem-1515 142 30 , , PUNCT fbem-1515 142 31 the the DET fbem-1515 142 32 first first ADJ fbem-1515 142 33 5 5 NUM fbem-1515 142 34 nodes node NOUN fbem-1515 142 35 in in ADP fbem-1515 142 36 decision decision NOUN fbem-1515 142 37 tree tree NOUN fbem-1515 142 38 are be AUX fbem-1515 142 39 ' ' PUNCT fbem-1515 142 40 score_source_3_z score_source_3_z PROPN fbem-1515 142 41 ' ' PUNCT fbem-1515 142 42 , , PUNCT fbem-1515 142 43 ' ' PUNCT fbem-1515 142 44 score_source_2 score_source_2 NOUN fbem-1515 142 45 ' ' NUM fbem-1515 142 46 , , PUNCT fbem-1515 142 47 ' ' PUNCT fbem-1515 142 48 score_source_1_z score_source_1_z ADV fbem-1515 142 49 ' ' PUNCT fbem-1515 142 50 , , PUNCT fbem-1515 142 51 ' ' PUNCT fbem-1515 142 52 employed_days_z employed_days_z NOUN fbem-1515 142 53 ' ' PUNCT fbem-1515 142 54 ( ( PUNCT fbem-1515 142 55 z z NOUN fbem-1515 142 56 score score NOUN fbem-1515 142 57 of of ADP fbem-1515 142 58 days day NOUN fbem-1515 142 59 before before ADP fbem-1515 142 60 the the DET fbem-1515 142 61 application application NOUN fbem-1515 142 62 , , PUNCT fbem-1515 142 63 the the DET fbem-1515 142 64 client client NOUN fbem-1515 142 65 started start VERB fbem-1515 142 66 earning earn VERB fbem-1515 142 67 ) ) PUNCT fbem-1515 142 68 , , PUNCT fbem-1515 142 69 ' ' PUNCT fbem-1515 142 70 client_education_secondary client_education_secondary ADJ fbem-1515 142 71 ' ' PUNCT fbem-1515 142 72 , , PUNCT fbem-1515 142 73 which which PRON fbem-1515 142 74 show show VERB fbem-1515 142 75 large large ADJ fbem-1515 142 76 similarity similarity NOUN fbem-1515 142 77 with with ADP fbem-1515 142 78 those those PRON fbem-1515 142 79 in in ADP fbem-1515 142 80 our our PRON fbem-1515 142 81 logistic logistic ADJ fbem-1515 142 82 model model NOUN fbem-1515 142 83 . . PUNCT fbem-1515 143 1 so so ADV fbem-1515 143 2 , , PUNCT fbem-1515 143 3 we we PRON fbem-1515 143 4 can can AUX fbem-1515 143 5 conclude conclude VERB fbem-1515 143 6 that that SCONJ fbem-1515 143 7 the the DET fbem-1515 143 8 features feature NOUN fbem-1515 143 9 of of ADP fbem-1515 143 10 score score NOUN fbem-1515 143 11 1 1 NUM fbem-1515 143 12 , , PUNCT fbem-1515 143 13 2 2 NUM fbem-1515 143 14 , , PUNCT fbem-1515 143 15 3 3 NUM fbem-1515 143 16 and and CCONJ fbem-1515 143 17 whether whether SCONJ fbem-1515 143 18 clients client NOUN fbem-1515 143 19 have have AUX fbem-1515 143 20 finished finish VERB fbem-1515 143 21 junior junior ADJ fbem-1515 143 22 secondary secondary ADJ fbem-1515 143 23 education education NOUN fbem-1515 143 24 are be AUX fbem-1515 143 25 key key ADJ fbem-1515 143 26 features feature NOUN fbem-1515 143 27 in in ADP fbem-1515 143 28 predicting predict VERB fbem-1515 143 29 default default NOUN fbem-1515 143 30 states state NOUN fbem-1515 143 31 . . PUNCT fbem-1515 144 1 and and CCONJ fbem-1515 144 2 we we PRON fbem-1515 144 3 need need VERB fbem-1515 144 4 to to PART fbem-1515 144 5 pay pay VERB fbem-1515 144 6 special special ADJ fbem-1515 144 7 attention attention NOUN fbem-1515 144 8 to to ADP fbem-1515 144 9 those those DET fbem-1515 144 10 features feature NOUN fbem-1515 144 11 when when SCONJ fbem-1515 144 12 collecting collect VERB fbem-1515 144 13 data datum NOUN fbem-1515 144 14 , , PUNCT fbem-1515 144 15 setting set VERB fbem-1515 144 16 those those DET fbem-1515 144 17 features feature NOUN fbem-1515 144 18 as as ADP fbem-1515 144 19 required require VERB fbem-1515 144 20 fields field NOUN fbem-1515 144 21 when when SCONJ fbem-1515 144 22 collecting collect VERB fbem-1515 144 23 the the DET fbem-1515 144 24 information information NOUN fbem-1515 144 25 of of ADP fbem-1515 144 26 clients client NOUN fbem-1515 144 27 during during ADP fbem-1515 144 28 their their PRON fbem-1515 144 29 application application NOUN fbem-1515 144 30 of of ADP fbem-1515 144 31 our our PRON fbem-1515 144 32 car car NOUN fbem-1515 144 33 loan loan NOUN fbem-1515 144 34 . . PUNCT fbem-1515 145 1 business business NOUN fbem-1515 145 2 strategy strategy NOUN fbem-1515 145 3 we we PRON fbem-1515 145 4 suggest suggest VERB fbem-1515 145 5 the the DET fbem-1515 145 6 financial financial ADJ fbem-1515 145 7 institution institution NOUN fbem-1515 145 8 to to PART fbem-1515 145 9 reject reject VERB fbem-1515 145 10 car car NOUN fbem-1515 145 11 loan loan NOUN fbem-1515 145 12 applications application NOUN fbem-1515 145 13 which which PRON fbem-1515 145 14 predict predict VERB fbem-1515 145 15 to to PART fbem-1515 145 16 be be AUX fbem-1515 145 17 β€˜ ' PUNCT fbem-1515 145 18 default default NOUN fbem-1515 145 19 ’ ' PUNCT fbem-1515 145 20 . . PUNCT fbem-1515 146 1 for for ADP fbem-1515 146 2 the the DET fbem-1515 146 3 car car NOUN fbem-1515 146 4 loan loan NOUN fbem-1515 146 5 application application NOUN fbem-1515 146 6 with with ADP fbem-1515 146 7 90%-95 90%-95 NUM fbem-1515 146 8 % % NOUN fbem-1515 146 9 probability probability NOUN fbem-1515 146 10 not not PART fbem-1515 146 11 to to PART fbem-1515 146 12 default default VERB fbem-1515 146 13 , , PUNCT fbem-1515 146 14 fi fi NOUN fbem-1515 146 15 can can AUX fbem-1515 146 16 consider consider VERB fbem-1515 146 17 set set VERB fbem-1515 146 18 limits limit NOUN fbem-1515 146 19 on on ADP fbem-1515 146 20 the the DET fbem-1515 146 21 loan loan NOUN fbem-1515 146 22 amount amount NOUN fbem-1515 146 23 . . PUNCT fbem-1515 147 1 after after ADP fbem-1515 147 2 the the DET fbem-1515 147 3 application application NOUN fbem-1515 147 4 stage stage NOUN fbem-1515 147 5 , , PUNCT fbem-1515 147 6 fi fi NOUN fbem-1515 147 7 can can AUX fbem-1515 147 8 consider consider VERB fbem-1515 147 9 set set VERB fbem-1515 147 10 a a DET fbem-1515 147 11 6 6 NUM fbem-1515 147 12 -12month -12month NUM fbem-1515 147 13 time time NOUN fbem-1515 147 14 window window NOUN fbem-1515 147 15 to to PART fbem-1515 147 16 observe observe VERB fbem-1515 147 17 the the DET fbem-1515 147 18 customer customer NOUN fbem-1515 147 19 behavior behavior NOUN fbem-1515 147 20 and and CCONJ fbem-1515 147 21 collect collect VERB fbem-1515 147 22 data datum NOUN fbem-1515 147 23 such such ADJ fbem-1515 147 24 as as ADP fbem-1515 147 25 β€˜ ' PUNCT fbem-1515 147 26 day day NOUN fbem-1515 147 27 past past ADJ fbem-1515 147 28 due due ADJ fbem-1515 147 29 ’ ' PUNCT fbem-1515 147 30 to to PART fbem-1515 147 31 adjust adjust VERB fbem-1515 147 32 the the DET fbem-1515 147 33 strategy strategy NOUN fbem-1515 147 34 accordingly accordingly ADV fbem-1515 147 35 . . PUNCT fbem-1515 148 1 7 7 X fbem-1515 148 2 . . X fbem-1515 148 3 reflections reflection NOUN fbem-1515 148 4 a a PRON fbem-1515 148 5 ) ) PUNCT fbem-1515 148 6 data datum NOUN fbem-1515 148 7 processing processing NOUN fbem-1515 148 8 i i NOUN fbem-1515 148 9 ) ) PUNCT fbem-1515 148 10 . . PUNCT fbem-1515 149 1 the the DET fbem-1515 149 2 proportion proportion NOUN fbem-1515 149 3 of of ADP fbem-1515 149 4 default default NOUN fbem-1515 149 5 examples example NOUN fbem-1515 149 6 : : PUNCT fbem-1515 149 7 as as SCONJ fbem-1515 149 8 previously previously ADV fbem-1515 149 9 discussed discuss VERB fbem-1515 149 10 in in ADP fbem-1515 149 11 this this DET fbem-1515 149 12 report report NOUN fbem-1515 149 13 , , PUNCT fbem-1515 149 14 we we PRON fbem-1515 149 15 constructed construct VERB fbem-1515 149 16 2 2 NUM fbem-1515 149 17 datasets dataset NOUN fbem-1515 149 18 , , PUNCT fbem-1515 149 19 one one NUM fbem-1515 149 20 with with ADP fbem-1515 149 21 all all DET fbem-1515 149 22 non non ADJ fbem-1515 149 23 - - NOUN fbem-1515 149 24 default default ADJ fbem-1515 149 25 na na ADP fbem-1515 149 26 examples example NOUN fbem-1515 149 27 dropped drop VERB fbem-1515 149 28 ( ( PUNCT fbem-1515 149 29 train_1 train_1 PROPN fbem-1515 149 30 ) ) PUNCT fbem-1515 149 31 while while SCONJ fbem-1515 149 32 the the DET fbem-1515 149 33 other other ADJ fbem-1515 149 34 dataset dataset NOUN fbem-1515 149 35 keeps keep VERB fbem-1515 149 36 all all DET fbem-1515 149 37 examples example NOUN fbem-1515 149 38 ( ( PUNCT fbem-1515 149 39 train train NOUN fbem-1515 149 40 ) ) PUNCT fbem-1515 149 41 . . PUNCT fbem-1515 150 1 in in ADP fbem-1515 150 2 our our PRON fbem-1515 150 3 original original ADJ fbem-1515 150 4 expectation expectation NOUN fbem-1515 150 5 , , PUNCT fbem-1515 150 6 we we PRON fbem-1515 150 7 thought think VERB fbem-1515 150 8 models model NOUN fbem-1515 150 9 built build VERB fbem-1515 150 10 on on ADP fbem-1515 150 11 train_1 train_1 NOUN fbem-1515 150 12 would would AUX fbem-1515 150 13 capture capture VERB fbem-1515 150 14 the the DET fbem-1515 150 15 essential essential ADJ fbem-1515 150 16 characteristics characteristic NOUN fbem-1515 150 17 of of ADP fbem-1515 150 18 default default NOUN fbem-1515 150 19 cases case NOUN fbem-1515 150 20 better well ADJ fbem-1515 150 21 , , PUNCT fbem-1515 150 22 which which PRON fbem-1515 150 23 we we PRON fbem-1515 150 24 care care VERB fbem-1515 150 25 about about ADP fbem-1515 150 26 the the DET fbem-1515 150 27 most most ADJ fbem-1515 150 28 for for ADP fbem-1515 150 29 profit profit NOUN fbem-1515 150 30 reasons reason NOUN fbem-1515 150 31 . . PUNCT fbem-1515 151 1 however however ADV fbem-1515 151 2 , , PUNCT fbem-1515 151 3 though though SCONJ fbem-1515 151 4 models model NOUN fbem-1515 151 5 built build VERB fbem-1515 151 6 on on ADP fbem-1515 151 7 train_1 train_1 NOUN fbem-1515 151 8 turn turn VERB fbem-1515 151 9 to to PART fbem-1515 151 10 have have VERB fbem-1515 151 11 higher high ADJ fbem-1515 151 12 auc auc NOUN fbem-1515 151 13 in in ADP fbem-1515 151 14 general general ADJ fbem-1515 151 15 , , PUNCT fbem-1515 151 16 they they PRON fbem-1515 151 17 perform perform VERB fbem-1515 151 18 too too ADV fbem-1515 151 19 poorly poorly ADV fbem-1515 151 20 in in ADP fbem-1515 151 21 terms term NOUN fbem-1515 151 22 of of ADP fbem-1515 151 23 accuracy accuracy NOUN fbem-1515 151 24 . . PUNCT fbem-1515 152 1 this this PRON fbem-1515 152 2 may may AUX fbem-1515 152 3 be be AUX fbem-1515 152 4 due due ADJ fbem-1515 152 5 to to ADP fbem-1515 152 6 the the DET fbem-1515 152 7 train_1 train_1 NOUN fbem-1515 152 8 dataset dataset VERB fbem-1515 152 9 having have VERB fbem-1515 152 10 a a DET fbem-1515 152 11 very very ADV fbem-1515 152 12 different different ADJ fbem-1515 152 13 proportion proportion NOUN fbem-1515 152 14 of of ADP fbem-1515 152 15 default default NOUN fbem-1515 152 16 nom nom NOUN fbem-1515 152 17 - - NOUN fbem-1515 152 18 default default NOUN fbem-1515 152 19 from from ADP fbem-1515 152 20 reality reality NOUN fbem-1515 152 21 , , PUNCT fbem-1515 152 22 so so SCONJ fbem-1515 152 23 there there PRON fbem-1515 152 24 would would AUX fbem-1515 152 25 be be AUX fbem-1515 152 26 accuracy accuracy NOUN fbem-1515 152 27 pitfalls pitfall NOUN fbem-1515 152 28 when when SCONJ fbem-1515 152 29 we we PRON fbem-1515 152 30 try try VERB fbem-1515 152 31 to to PART fbem-1515 152 32 use use VERB fbem-1515 152 33 the the DET fbem-1515 152 34 model model NOUN fbem-1515 152 35 to to PART fbem-1515 152 36 predict predict VERB fbem-1515 152 37 some some DET fbem-1515 152 38 real real ADJ fbem-1515 152 39 data datum NOUN fbem-1515 152 40 . . PUNCT fbem-1515 152 41 ii ii PROPN fbem-1515 152 42 ) ) PUNCT fbem-1515 152 43 . . PUNCT fbem-1515 153 1 possible possible ADJ fbem-1515 153 2 interaction interaction NOUN fbem-1515 153 3 between between ADP fbem-1515 153 4 features feature NOUN fbem-1515 153 5 unknown unknown ADJ fbem-1515 153 6 : : PUNCT fbem-1515 153 7 there there PRON fbem-1515 153 8 may may AUX fbem-1515 153 9 exist exist VERB fbem-1515 153 10 some some DET fbem-1515 153 11 interactional interactional ADJ fbem-1515 153 12 relationships relationship NOUN fbem-1515 153 13 between between ADP fbem-1515 153 14 features feature NOUN fbem-1515 153 15 , , PUNCT fbem-1515 153 16 which which PRON fbem-1515 153 17 we we PRON fbem-1515 153 18 currently currently ADV fbem-1515 153 19 are be AUX fbem-1515 153 20 unaware unaware ADJ fbem-1515 153 21 of of ADP fbem-1515 153 22 . . PUNCT fbem-1515 154 1 exploration exploration NOUN fbem-1515 154 2 and and CCONJ fbem-1515 154 3 utilization utilization NOUN fbem-1515 154 4 of of ADP fbem-1515 154 5 those those DET fbem-1515 154 6 relationships relationship NOUN fbem-1515 154 7 may may AUX fbem-1515 154 8 be be AUX fbem-1515 154 9 beneficial beneficial ADJ fbem-1515 154 10 to to ADP fbem-1515 154 11 model model NOUN fbem-1515 154 12 building building NOUN fbem-1515 154 13 . . PUNCT fbem-1515 155 1 b b X fbem-1515 155 2 ) ) PUNCT fbem-1515 155 3 knn knn PROPN fbem-1515 155 4 model model NOUN fbem-1515 155 5 in in ADP fbem-1515 155 6 practice practice NOUN fbem-1515 155 7 though though SCONJ fbem-1515 155 8 individually individually ADV fbem-1515 155 9 , , PUNCT fbem-1515 155 10 the the DET fbem-1515 155 11 kkn kkn PROPN fbem-1515 155 12 model model NOUN fbem-1515 155 13 performed perform VERB fbem-1515 155 14 the the DET fbem-1515 155 15 best good ADJ fbem-1515 155 16 among among ADP fbem-1515 155 17 all all DET fbem-1515 155 18 the the DET fbem-1515 155 19 models model NOUN fbem-1515 155 20 in in ADP fbem-1515 155 21 terms term NOUN fbem-1515 155 22 of of ADP fbem-1515 155 23 auc auc NOUN fbem-1515 155 24 . . PUNCT fbem-1515 156 1 however however ADV fbem-1515 156 2 , , PUNCT fbem-1515 156 3 it it PRON fbem-1515 156 4 may may AUX fbem-1515 156 5 not not PART fbem-1515 156 6 be be AUX fbem-1515 156 7 very very ADV fbem-1515 156 8 practical practical ADJ fbem-1515 156 9 for for SCONJ fbem-1515 156 10 financial financial ADJ fbem-1515 156 11 institutions institution NOUN fbem-1515 156 12 to to PART fbem-1515 156 13 use use VERB fbem-1515 156 14 in in ADP fbem-1515 156 15 real real ADJ fbem-1515 156 16 life life NOUN fbem-1515 156 17 , , PUNCT fbem-1515 156 18 as as SCONJ fbem-1515 156 19 it it PRON fbem-1515 156 20 ’s ’ VERB fbem-1515 156 21 time time NOUN fbem-1515 156 22 - - PUNCT fbem-1515 156 23 consuming consume VERB fbem-1515 156 24 and and CCONJ fbem-1515 156 25 takes take VERB fbem-1515 156 26 up up ADP fbem-1515 156 27 too too ADV fbem-1515 156 28 much much ADJ fbem-1515 156 29 computational computational ADJ fbem-1515 156 30 power power NOUN fbem-1515 156 31 when when SCONJ fbem-1515 156 32 the the DET fbem-1515 156 33 volume volume NOUN fbem-1515 156 34 of of ADP fbem-1515 156 35 data datum NOUN fbem-1515 156 36 to to PART fbem-1515 156 37 be be AUX fbem-1515 156 38 processed process VERB fbem-1515 156 39 is be AUX fbem-1515 156 40 huge huge ADJ fbem-1515 156 41 . . PUNCT fbem-1515 157 1 c c X fbem-1515 157 2 ) ) PUNCT fbem-1515 157 3 underlying underlie VERB fbem-1515 157 4 logic logic NOUN fbem-1515 157 5 and and CCONJ fbem-1515 157 6 reasons reason NOUN fbem-1515 157 7 of of ADP fbem-1515 157 8 feature feature NOUN fbem-1515 157 9 importance importance NOUN fbem-1515 157 10 we we PRON fbem-1515 157 11 can can AUX fbem-1515 157 12 only only ADV fbem-1515 157 13 have have VERB fbem-1515 157 14 importance importance NOUN fbem-1515 157 15 scores score NOUN fbem-1515 157 16 of of ADP fbem-1515 157 17 features feature NOUN fbem-1515 157 18 in in ADP fbem-1515 157 19 the the DET fbem-1515 157 20 decision decision NOUN fbem-1515 157 21 tree tree NOUN fbem-1515 157 22 and and CCONJ fbem-1515 157 23 logistic logistic ADJ fbem-1515 157 24 models model NOUN fbem-1515 157 25 , , PUNCT fbem-1515 157 26 but but CCONJ fbem-1515 157 27 we we PRON fbem-1515 157 28 are be AUX fbem-1515 157 29 unaware unaware ADJ fbem-1515 157 30 of of ADP fbem-1515 157 31 the the DET fbem-1515 157 32 underlying underlie VERB fbem-1515 157 33 logic logic NOUN fbem-1515 157 34 and and CCONJ fbem-1515 157 35 reason reason NOUN fbem-1515 157 36 of of ADP fbem-1515 157 37 a a DET fbem-1515 157 38 feature feature NOUN fbem-1515 157 39 being be AUX fbem-1515 157 40 important important ADJ fbem-1515 157 41 . . PUNCT fbem-1515 158 1 for for ADP fbem-1515 158 2 example example NOUN fbem-1515 158 3 , , PUNCT fbem-1515 158 4 ' ' PUNCT fbem-1515 158 5 mobile_tag mobile_tag NOUN fbem-1515 158 6 ' ' PUNCT fbem-1515 158 7 ( ( PUNCT fbem-1515 158 8 mobile mobile ADJ fbem-1515 158 9 number number NOUN fbem-1515 158 10 provided provide VERB fbem-1515 158 11 by by ADP fbem-1515 158 12 client client NOUN fbem-1515 158 13 , , PUNCT fbem-1515 158 14 1 1 NUM fbem-1515 158 15 means mean VERB fbem-1515 158 16 yes yes INTJ fbem-1515 158 17 and and CCONJ fbem-1515 158 18 0 0 NUM fbem-1515 158 19 means mean VERB fbem-1515 158 20 no no NOUN fbem-1515 158 21 ) ) PUNCT fbem-1515 158 22 feature feature NOUN fbem-1515 158 23 ranks rank VERB fbem-1515 158 24 top top NOUN fbem-1515 158 25 among among ADP fbem-1515 158 26 149 149 NUM fbem-1515 158 27 logistic logistic ADJ fbem-1515 158 28 model model NOUN fbem-1515 158 29 features feature NOUN fbem-1515 158 30 . . PUNCT fbem-1515 159 1 however however ADV fbem-1515 159 2 , , PUNCT fbem-1515 159 3 the the DET fbem-1515 159 4 real real ADJ fbem-1515 159 5 reason reason NOUN fbem-1515 159 6 for for ADP fbem-1515 159 7 ' ' PUNCT fbem-1515 159 8 mobile_tag mobile_tag NOUN fbem-1515 159 9 ' ' PUNCT fbem-1515 159 10 having have VERB fbem-1515 159 11 high high ADJ fbem-1515 159 12 predictivity predictivity NOUN fbem-1515 159 13 could could AUX fbem-1515 159 14 possibly possibly ADV fbem-1515 159 15 be be AUX fbem-1515 159 16 that that SCONJ fbem-1515 159 17 whether whether SCONJ fbem-1515 159 18 a a DET fbem-1515 159 19 mobile mobile ADJ fbem-1515 159 20 phone phone NOUN fbem-1515 159 21 number number NOUN fbem-1515 159 22 can can AUX fbem-1515 159 23 be be AUX fbem-1515 159 24 provided provide VERB fbem-1515 159 25 may may AUX fbem-1515 159 26 reflect reflect VERB fbem-1515 159 27 the the DET fbem-1515 159 28 stability stability NOUN fbem-1515 159 29 and and CCONJ fbem-1515 159 30 financial financial ADJ fbem-1515 159 31 state state NOUN fbem-1515 159 32 of of ADP fbem-1515 159 33 a a DET fbem-1515 159 34 customer customer NOUN fbem-1515 159 35 . . PUNCT fbem-1515 160 1 then then ADV fbem-1515 160 2 we we PRON fbem-1515 160 3 may may AUX fbem-1515 160 4 also also ADV fbem-1515 160 5 try try VERB fbem-1515 160 6 to to PART fbem-1515 160 7 collect collect VERB fbem-1515 160 8 more more ADJ fbem-1515 160 9 information information NOUN fbem-1515 160 10 about about ADP fbem-1515 160 11 the the DET fbem-1515 160 12 stability stability NOUN fbem-1515 160 13 of of ADP fbem-1515 160 14 a a DET fbem-1515 160 15 customer customer NOUN fbem-1515 160 16 applying apply VERB fbem-1515 160 17 for for ADP fbem-1515 160 18 a a DET fbem-1515 160 19 loan loan NOUN fbem-1515 160 20 . . PUNCT fbem-1515 161 1 if if SCONJ fbem-1515 161 2 we we PRON fbem-1515 161 3 stretch stretch VERB fbem-1515 161 4 further far ADV fbem-1515 161 5 to to ADP fbem-1515 161 6 those those DET fbem-1515 161 7 underlying underlying ADJ fbem-1515 161 8 logic logic NOUN fbem-1515 161 9 and and CCONJ fbem-1515 161 10 reasons reason NOUN fbem-1515 161 11 ; ; PUNCT fbem-1515 161 12 we we PRON fbem-1515 161 13 may may AUX fbem-1515 161 14 collect collect VERB fbem-1515 161 15 data datum NOUN fbem-1515 161 16 more more ADV fbem-1515 161 17 effectively effectively ADV fbem-1515 161 18 to to PART fbem-1515 161 19 achieve achieve VERB fbem-1515 161 20 better well ADJ fbem-1515 161 21 prediction prediction NOUN fbem-1515 161 22 results result NOUN fbem-1515 161 23 . . PUNCT fbem-1515 162 1 d d X fbem-1515 162 2 ) ) PUNCT fbem-1515 162 3 sensitivity sensitivity NOUN fbem-1515 162 4 to to PART fbem-1515 162 5 change change VERB fbem-1515 162 6 of of ADP fbem-1515 162 7 cost cost NOUN fbem-1515 162 8 analysis analysis NOUN fbem-1515 162 9 considering consider VERB fbem-1515 162 10 the the DET fbem-1515 162 11 practice practice NOUN fbem-1515 162 12 of of ADP fbem-1515 162 13 selecting select VERB fbem-1515 162 14 an an DET fbem-1515 162 15 optimal optimal ADJ fbem-1515 162 16 decision decision NOUN fbem-1515 162 17 threshold threshold NOUN fbem-1515 162 18 based base VERB fbem-1515 162 19 on on ADP fbem-1515 162 20 the the DET fbem-1515 162 21 cost cost NOUN fbem-1515 162 22 analysis analysis NOUN fbem-1515 162 23 , , PUNCT fbem-1515 162 24 the the DET fbem-1515 162 25 sensitivity sensitivity NOUN fbem-1515 162 26 to to PART fbem-1515 162 27 change change VERB fbem-1515 162 28 of of ADP fbem-1515 162 29 the the DET fbem-1515 162 30 cost cost NOUN fbem-1515 162 31 analysis analysis NOUN fbem-1515 162 32 should should AUX fbem-1515 162 33 be be AUX fbem-1515 162 34 considered consider VERB fbem-1515 162 35 . . PUNCT fbem-1515 163 1 changes change NOUN fbem-1515 163 2 in in ADP fbem-1515 163 3 parameters parameter NOUN fbem-1515 163 4 like like ADP fbem-1515 163 5 average average ADJ fbem-1515 163 6 exposure exposure NOUN fbem-1515 163 7 , , PUNCT fbem-1515 163 8 commission commission NOUN fbem-1515 163 9 rate rate NOUN fbem-1515 163 10 , , PUNCT fbem-1515 163 11 etc etc X fbem-1515 163 12 . . X fbem-1515 163 13 , , PUNCT fbem-1515 163 14 may may AUX fbem-1515 163 15 affect affect VERB fbem-1515 163 16 the the DET fbem-1515 163 17 result result NOUN fbem-1515 163 18 of of ADP fbem-1515 163 19 the the DET fbem-1515 163 20 analysis analysis NOUN fbem-1515 163 21 , , PUNCT fbem-1515 163 22 changing change VERB fbem-1515 163 23 the the DET fbem-1515 163 24 optimal optimal ADJ fbem-1515 163 25 model model NOUN fbem-1515 163 26 . . PUNCT fbem-1515 164 1 8 8 X fbem-1515 164 2 . . X fbem-1515 164 3 summary summary NOUN fbem-1515 164 4 according accord VERB fbem-1515 164 5 to to ADP fbem-1515 164 6 the the DET fbem-1515 164 7 characteristics characteristic NOUN fbem-1515 164 8 of of ADP fbem-1515 164 9 our our PRON fbem-1515 164 10 data datum NOUN fbem-1515 164 11 ( ( PUNCT fbem-1515 164 12 large large ADJ fbem-1515 164 13 size size NOUN fbem-1515 164 14 with with ADP fbem-1515 164 15 a a DET fbem-1515 164 16 great great ADJ fbem-1515 164 17 number number NOUN fbem-1515 164 18 of of ADP fbem-1515 164 19 features feature NOUN fbem-1515 164 20 ; ; PUNCT fbem-1515 164 21 in in ADV fbem-1515 164 22 - - PUNCT fbem-1515 164 23 balanced balanced ADJ fbem-1515 164 24 in in ADP fbem-1515 164 25 general general ADJ fbem-1515 164 26 ) ) PUNCT fbem-1515 164 27 , , PUNCT fbem-1515 164 28 we we PRON fbem-1515 164 29 applied apply VERB fbem-1515 164 30 2 2 NUM fbem-1515 164 31 different different ADJ fbem-1515 164 32 data datum NOUN fbem-1515 164 33 processing processing NOUN fbem-1515 164 34 methods method NOUN fbem-1515 164 35 ( ( PUNCT fbem-1515 164 36 dropping drop VERB fbem-1515 164 37 nondefault nondefault NOUN fbem-1515 164 38 nas nas ADV fbem-1515 164 39 or or CCONJ fbem-1515 164 40 keeping keep VERB fbem-1515 164 41 them they PRON fbem-1515 164 42 ) ) PUNCT fbem-1515 164 43 and and CCONJ fbem-1515 164 44 got get VERB fbem-1515 164 45 2 2 NUM fbem-1515 164 46 datasets dataset NOUN fbem-1515 164 47 . . PUNCT fbem-1515 165 1 and and CCONJ fbem-1515 165 2 with with ADP fbem-1515 165 3 our our PRON fbem-1515 165 4 existing exist VERB fbem-1515 165 5 skill skill NOUN fbem-1515 165 6 sets set NOUN fbem-1515 165 7 , , PUNCT fbem-1515 165 8 we we PRON fbem-1515 165 9 built build VERB fbem-1515 165 10 the the DET fbem-1515 165 11 following follow VERB fbem-1515 165 12 individual individual ADJ fbem-1515 165 13 models model NOUN fbem-1515 165 14 . . PUNCT fbem-1515 166 1 and and CCONJ fbem-1515 166 2 selected select VERB fbem-1515 166 3 3 3 NUM fbem-1515 166 4 models model NOUN fbem-1515 166 5 to to PART fbem-1515 166 6 be be AUX fbem-1515 166 7 adopted adopt VERB fbem-1515 166 8 based base VERB fbem-1515 166 9 on on ADP fbem-1515 166 10 their their PRON fbem-1515 166 11 performance performance NOUN fbem-1515 166 12 in in ADP fbem-1515 166 13 terms term NOUN fbem-1515 166 14 of of ADP fbem-1515 166 15 accuracy accuracy NOUN fbem-1515 166 16 , , PUNCT fbem-1515 166 17 confusion confusion NOUN fbem-1515 166 18 metrix metrix NOUN fbem-1515 166 19 , , PUNCT fbem-1515 166 20 and and CCONJ fbem-1515 166 21 auc auc NOUN fbem-1515 166 22 . . PUNCT fbem-1515 167 1 model model NOUN fbem-1515 167 2 type type NOUN fbem-1515 167 3 data datum NOUN fbem-1515 167 4 set set VERB fbem-1515 167 5 used use VERB fbem-1515 167 6 adopted adopt VERB fbem-1515 167 7 decision decision NOUN fbem-1515 167 8 tree tree NOUN fbem-1515 167 9 model model NOUN fbem-1515 167 10 train train NOUN fbem-1515 168 1 yes yes INTJ fbem-1515 168 2 train_1 train_1 NOUN fbem-1515 168 3 no no DET fbem-1515 168 4 logistic logistic ADJ fbem-1515 168 5 model model NOUN fbem-1515 168 6 train train NOUN fbem-1515 169 1 yes yes INTJ fbem-1515 169 2 train_1 train_1 INTJ fbem-1515 169 3 no no DET fbem-1515 169 4 knn knn NOUN fbem-1515 169 5 model model PROPN fbem-1515 169 6 train train NOUN fbem-1515 170 1 yes yes INTJ fbem-1515 170 2 train_1 train_1 VERB fbem-1515 170 3 no no PRON fbem-1515 170 4 keeping keep VERB fbem-1515 170 5 our our PRON fbem-1515 170 6 target target NOUN fbem-1515 170 7 of of ADP fbem-1515 170 8 making make VERB fbem-1515 170 9 predictions prediction NOUN fbem-1515 170 10 that that PRON fbem-1515 170 11 maximize maximize VERB fbem-1515 170 12 profits profit NOUN fbem-1515 170 13 in in ADP fbem-1515 170 14 mind mind NOUN fbem-1515 170 15 , , PUNCT fbem-1515 170 16 we we PRON fbem-1515 170 17 did do AUX fbem-1515 170 18 cost cost VERB fbem-1515 170 19 analysis analysis NOUN fbem-1515 170 20 and and CCONJ fbem-1515 170 21 adjusted adjust VERB fbem-1515 170 22 the the DET fbem-1515 170 23 thresholds threshold NOUN fbem-1515 170 24 to to PART fbem-1515 170 25 be be AUX fbem-1515 170 26 optimal optimal ADJ fbem-1515 170 27 in in ADP fbem-1515 170 28 each each DET fbem-1515 170 29 adopted adopt VERB fbem-1515 170 30 model model NOUN fbem-1515 170 31 . . PUNCT fbem-1515 171 1 according accord VERB fbem-1515 171 2 to to ADP fbem-1515 171 3 the the DET fbem-1515 171 4 feature feature NOUN fbem-1515 171 5 importance importance NOUN fbem-1515 171 6 ranking ranking NOUN fbem-1515 171 7 of of ADP fbem-1515 171 8 decision decision NOUN fbem-1515 171 9 tree tree NOUN fbem-1515 171 10 model model NOUN fbem-1515 171 11 and and CCONJ fbem-1515 171 12 logistic logistic ADJ fbem-1515 171 13 model model NOUN fbem-1515 171 14 , , PUNCT fbem-1515 171 15 we we PRON fbem-1515 171 16 can can AUX fbem-1515 171 17 conclude conclude VERB fbem-1515 171 18 that that SCONJ fbem-1515 171 19 the the DET fbem-1515 171 20 features feature NOUN fbem-1515 171 21 of of ADP fbem-1515 171 22 score score NOUN fbem-1515 171 23 1 1 NUM fbem-1515 171 24 , , PUNCT fbem-1515 171 25 2 2 NUM fbem-1515 171 26 , , PUNCT fbem-1515 171 27 3 3 NUM fbem-1515 171 28 and and CCONJ fbem-1515 171 29 whether whether SCONJ fbem-1515 171 30 clients client NOUN fbem-1515 171 31 have have AUX fbem-1515 171 32 finished finish VERB fbem-1515 171 33 junior junior ADJ fbem-1515 171 34 secondary secondary ADJ fbem-1515 171 35 education education NOUN fbem-1515 171 36 have have VERB fbem-1515 171 37 the the DET fbem-1515 171 38 highest high ADJ fbem-1515 171 39 predictivity predictivity NOUN fbem-1515 171 40 in in ADP fbem-1515 171 41 the the DET fbem-1515 171 42 loan loan NOUN fbem-1515 171 43 default default NOUN fbem-1515 171 44 results result NOUN fbem-1515 171 45 . . PUNCT fbem-1515 172 1 and and CCONJ fbem-1515 172 2 we we PRON fbem-1515 172 3 suggest suggest VERB fbem-1515 172 4 the the DET fbem-1515 172 5 institution institution NOUN fbem-1515 172 6 outsources outsource NOUN fbem-1515 172 7 and and CCONJ fbem-1515 172 8 pays pay VERB fbem-1515 172 9 more more ADJ fbem-1515 172 10 attention attention NOUN fbem-1515 172 11 to to ADP fbem-1515 172 12 the the DET fbem-1515 172 13 above above ADJ fbem-1515 172 14 4 4 NUM fbem-1515 172 15 features feature NOUN fbem-1515 172 16 in in ADP fbem-1515 172 17 terms term NOUN fbem-1515 172 18 of of ADP fbem-1515 172 19 data data NOUN fbem-1515 172 20 collection collection NOUN fbem-1515 172 21 . . PUNCT fbem-1515 173 1 and and CCONJ fbem-1515 173 2 we we PRON fbem-1515 173 3 stretched stretch VERB fbem-1515 173 4 further far ADV fbem-1515 173 5 to to PART fbem-1515 173 6 incorporate incorporate VERB fbem-1515 173 7 the the DET fbem-1515 173 8 3 3 NUM fbem-1515 173 9 models model NOUN fbem-1515 173 10 and and CCONJ fbem-1515 173 11 tried try VERB fbem-1515 173 12 2 2 NUM fbem-1515 173 13 other other ADJ fbem-1515 173 14 new new ADJ fbem-1515 173 15 models model NOUN fbem-1515 173 16 ( ( PUNCT fbem-1515 173 17 random random ADJ fbem-1515 173 18 forest forest NOUN fbem-1515 173 19 and and CCONJ fbem-1515 173 20 adaptive adaptive ADJ fbem-1515 173 21 oost oost NOUN fbem-1515 173 22 ) ) PUNCT fbem-1515 173 23 to to PART fbem-1515 173 24 build build VERB fbem-1515 173 25 ensemble ensemble ADJ fbem-1515 173 26 model model NOUN fbem-1515 173 27 with with ADP fbem-1515 173 28 different different ADJ fbem-1515 173 29 combinations combination NOUN fbem-1515 173 30 . . PUNCT fbem-1515 174 1 and and CCONJ fbem-1515 174 2 we we PRON fbem-1515 174 3 also also ADV fbem-1515 174 4 tried try VERB fbem-1515 174 5 out out ADP fbem-1515 174 6 2 2 NUM fbem-1515 174 7 rules rule NOUN fbem-1515 174 8 , , PUNCT fbem-1515 174 9 which which PRON fbem-1515 174 10 are be AUX fbem-1515 174 11 β€˜ ' PUNCT fbem-1515 174 12 predict predict VERB fbem-1515 174 13 default default NOUN fbem-1515 174 14 when when SCONJ fbem-1515 174 15 majority majority NOUN fbem-1515 174 16 predicts predict VERB fbem-1515 174 17 default default NOUN fbem-1515 174 18 ’ ' PUNCT fbem-1515 174 19 and and CCONJ fbem-1515 174 20 β€˜ ' PUNCT fbem-1515 174 21 predict predict VERB fbem-1515 174 22 default default NOUN fbem-1515 174 23 when when SCONJ fbem-1515 174 24 all all PRON fbem-1515 174 25 predict predict VERB fbem-1515 174 26 default default NOUN fbem-1515 174 27 ’ ' PUNCT fbem-1515 174 28 . . PUNCT fbem-1515 175 1 eventually eventually ADV fbem-1515 175 2 , , PUNCT fbem-1515 175 3 we we PRON fbem-1515 175 4 decide decide VERB fbem-1515 175 5 that that DET fbem-1515 175 6 ensemble ensemble ADJ fbem-1515 175 7 model model NOUN fbem-1515 175 8 ( ( PUNCT fbem-1515 175 9 knn knn PROPN fbem-1515 175 10 , , PUNCT fbem-1515 175 11 logistic logistic ADJ fbem-1515 175 12 and and CCONJ fbem-1515 175 13 random random ADJ fbem-1515 175 14 forest forest NOUN fbem-1515 175 15 ) ) PUNCT fbem-1515 175 16 with with ADP fbem-1515 175 17 majority majority NOUN fbem-1515 175 18 prediction prediction NOUN fbem-1515 175 19 and and CCONJ fbem-1515 175 20 0.9 0.9 NUM fbem-1515 175 21 threshold threshold NOUN fbem-1515 175 22 performs perform VERB fbem-1515 175 23 the the DET fbem-1515 175 24 best good ADJ fbem-1515 175 25 . . PUNCT fbem-1515 176 1 references reference NOUN fbem-1515 176 2 [ [ X fbem-1515 176 3 1 1 X fbem-1515 176 4 ] ] PUNCT fbem-1515 176 5 wang wang PROPN fbem-1515 176 6 p p PROPN fbem-1515 176 7 , , PUNCT fbem-1515 176 8 zheng zheng PROPN fbem-1515 176 9 h h PROPN fbem-1515 176 10 , , PUNCT fbem-1515 176 11 chen chen PROPN fbem-1515 176 12 d d PROPN fbem-1515 176 13 , , PUNCT fbem-1515 176 14 et et PROPN fbem-1515 176 15 al al PROPN fbem-1515 176 16 . . PUNCT fbem-1515 176 17 exploring explore VERB fbem-1515 176 18 the the DET fbem-1515 176 19 critical critical ADJ fbem-1515 176 20 factors factor NOUN fbem-1515 176 21 influencing influence VERB fbem-1515 176 22 online online ADJ fbem-1515 176 23 lending lending NOUN fbem-1515 176 24 intentions[j intentions[j PROPN fbem-1515 176 25 ] ] PUNCT fbem-1515 176 26 . . PUNCT fbem-1515 177 1 financial financial ADJ fbem-1515 177 2 innovation innovation NOUN fbem-1515 177 3 , , PUNCT fbem-1515 177 4 2015 2015 NUM fbem-1515 177 5 , , PUNCT fbem-1515 177 6 1(1):1 1(1):1 PROPN fbem-1515 177 7 - - NOUN fbem-1515 177 8 11 11 NUM fbem-1515 177 9 . . PUNCT fbem-1515 178 1 [ [ X fbem-1515 178 2 2 2 X fbem-1515 178 3 ] ] X fbem-1515 178 4 galindo galindo PROPN fbem-1515 178 5 j j PROPN fbem-1515 178 6 , , PUNCT fbem-1515 178 7 tamayo tamayo PROPN fbem-1515 178 8 p p PROPN fbem-1515 178 9 . . PUNCT fbem-1515 179 1 credit credit NOUN fbem-1515 179 2 risk risk NOUN fbem-1515 179 3 assessment assessment NOUN fbem-1515 179 4 using use VERB fbem-1515 179 5 statistical statistical ADJ fbem-1515 179 6 and and CCONJ fbem-1515 179 7 machine machine NOUN fbem-1515 179 8 learning[j learning[j NOUN fbem-1515 179 9 ] ] PUNCT fbem-1515 179 10 . . PUNCT fbem-1515 180 1 computational computational ADJ fbem-1515 180 2 economics economic NOUN fbem-1515 180 3 , , PUNCT fbem-1515 180 4 2000 2000 NUM fbem-1515 180 5 . . PUNCT fbem-1515 181 1 [ [ X fbem-1515 181 2 3 3 X fbem-1515 181 3 ] ] X fbem-1515 181 4 malekipirbazari malekipirbazari X fbem-1515 181 5 m m PROPN fbem-1515 181 6 , , PUNCT fbem-1515 181 7 aksakalli aksakalli PROPN fbem-1515 181 8 v v NOUN fbem-1515 181 9 . . PUNCT fbem-1515 182 1 risk risk NOUN fbem-1515 182 2 assessment assessment NOUN fbem-1515 182 3 in in ADP fbem-1515 182 4 social social ADJ fbem-1515 182 5 lending lending NOUN fbem-1515 182 6 via via ADP fbem-1515 182 7 random random ADJ fbem-1515 182 8 forests[j forests[j PROPN fbem-1515 182 9 ] ] PUNCT fbem-1515 182 10 . . PUNCT fbem-1515 183 1 expert expert NOUN fbem-1515 183 2 systems system NOUN fbem-1515 183 3 with with ADP fbem-1515 183 4 applications application NOUN fbem-1515 183 5 , , PUNCT fbem-1515 183 6 2015 2015 NUM fbem-1515 183 7 , , PUNCT fbem-1515 183 8 42(10):4621 42(10):4621 NUM fbem-1515 183 9 - - SYM fbem-1515 183 10 4631 4631 NUM fbem-1515 183 11 . . PUNCT fbem-1515 184 1 [ [ X fbem-1515 184 2 4 4 NUM fbem-1515 184 3 ] ] SYM fbem-1515 184 4 li li PROPN fbem-1515 184 5 y y PROPN fbem-1515 184 6 , , PUNCT fbem-1515 184 7 chen chen PROPN fbem-1515 184 8 w. w. PROPN fbem-1515 184 9 entropy entropy PROPN fbem-1515 184 10 method method NOUN fbem-1515 184 11 of of ADP fbem-1515 184 12 constructing construct VERB fbem-1515 184 13 a a DET fbem-1515 184 14 combined combined ADJ fbem-1515 184 15 model model NOUN fbem-1515 184 16 for for ADP fbem-1515 184 17 improving improve VERB fbem-1515 184 18 loan loan NOUN fbem-1515 184 19 default default NOUN fbem-1515 184 20 prediction prediction NOUN fbem-1515 184 21 : : PUNCT fbem-1515 184 22 a a DET fbem-1515 184 23 case case NOUN fbem-1515 184 24 study study NOUN fbem-1515 184 25 in in ADP fbem-1515 184 26 china[j china[j PROPN fbem-1515 184 27 ] ] PUNCT fbem-1515 184 28 . . PUNCT fbem-1515 185 1 journal journal PROPN fbem-1515 185 2 of of ADP fbem-1515 185 3 the the DET fbem-1515 185 4 operational operational ADJ fbem-1515 185 5 research research NOUN fbem-1515 185 6 society society NOUN fbem-1515 185 7 , , PUNCT fbem-1515 185 8 2019(4):1 2019(4):1 PROPN fbem-1515 185 9 - - SYM fbem-1515 185 10 11 11 NUM fbem-1515 185 11 . . PUNCT fbem-1515 186 1 [ [ X fbem-1515 186 2 5 5 NUM fbem-1515 186 3 ] ] X fbem-1515 186 4 zhou zhou PROPN fbem-1515 186 5 , , PUNCT fbem-1515 186 6 j. j. PROPN fbem-1515 186 7 , , PUNCT fbem-1515 186 8 li li PROPN fbem-1515 186 9 , , PUNCT fbem-1515 186 10 w. w. PROPN fbem-1515 186 11 , , PUNCT fbem-1515 186 12 wang wang PROPN fbem-1515 186 13 , , PUNCT fbem-1515 186 14 j. j. PROPN fbem-1515 186 15 , , PUNCT fbem-1515 186 16 ding ding PROPN fbem-1515 186 17 , , PUNCT fbem-1515 186 18 s. s. PROPN fbem-1515 186 19 , , PUNCT fbem-1515 186 20 xia xia PROPN fbem-1515 186 21 , , PUNCT fbem-1515 186 22 c. c. PROPN fbem-1515 186 23 , , PUNCT fbem-1515 186 24 2019 2019 NUM fbem-1515 186 25 . . PUNCT fbem-1515 186 26 default default NOUN fbem-1515 186 27 prediction prediction NOUN fbem-1515 186 28 in in ADP fbem-1515 186 29 p2p p2p NOUN fbem-1515 186 30 lending lend VERB fbem-1515 186 31 from from ADP fbem-1515 186 32 high high ADJ fbem-1515 186 33 - - PUNCT fbem-1515 186 34 dimensional dimensional ADJ fbem-1515 186 35 data datum NOUN fbem-1515 186 36 based base VERB fbem-1515 186 37 on on ADP fbem-1515 186 38 machine machine NOUN fbem-1515 186 39 learning learning NOUN fbem-1515 186 40 . . PUNCT fbem-1515 187 1 physica physica VERB fbem-1515 187 2 a a DET fbem-1515 187 3 statistical statistical ADJ fbem-1515 187 4 mechanics mechanic NOUN fbem-1515 187 5 and and CCONJ fbem-1515 187 6 its its PRON fbem-1515 187 7 applications application NOUN fbem-1515 187 8 534 534 NUM fbem-1515 187 9 , , PUNCT fbem-1515 187 10 122370 122370 NUM fbem-1515 187 11 . . PUNCT fbem-1515 188 1 doi doi NOUN fbem-1515 188 2 : : PUNCT fbem-1515 188 3 10.1016 10.1016 NUM fbem-1515 188 4 / / SYM fbem-1515 188 5 j.physa.2019.122370 j.physa.2019.122370 NOUN fbem-1515 188 6 . . PUNCT fbem-1515 189 1 [ [ X fbem-1515 189 2 6 6 NUM fbem-1515 189 3 ] ] X fbem-1515 189 4 zurada zurada PROPN fbem-1515 189 5 j. j. PROPN fbem-1515 189 6 data data PROPN fbem-1515 189 7 mining mining NOUN fbem-1515 189 8 techniques technique NOUN fbem-1515 189 9 in in ADP fbem-1515 189 10 predicting predict VERB fbem-1515 189 11 default default NOUN fbem-1515 189 12 rates rate NOUN fbem-1515 189 13 on on ADP fbem-1515 189 14 customer customer NOUN fbem-1515 189 15 loans[m]// loans[m]// PROPN fbem-1515 189 16 databases database NOUN fbem-1515 189 17 and and CCONJ fbem-1515 189 18 information information NOUN fbem-1515 189 19 systems systems PROPN fbem-1515 189 20 ii ii PROPN fbem-1515 189 21 . . PUNCT fbem-1515 189 22 springer springer PROPN fbem-1515 189 23 netherlands netherlands PROPN fbem-1515 189 24 , , PUNCT fbem-1515 189 25 2002 2002 NUM fbem-1515 189 26 . . PUNCT fbem-1515 190 1 [ [ X fbem-1515 190 2 7 7 NUM fbem-1515 190 3 ] ] PUNCT fbem-1515 190 4 breiman breiman NOUN fbem-1515 190 5 , , PUNCT fbem-1515 190 6 l. l. PROPN fbem-1515 190 7 ( ( PUNCT fbem-1515 190 8 2001 2001 NUM fbem-1515 190 9 ) ) PUNCT fbem-1515 190 10 . . PUNCT fbem-1515 191 1 random random ADJ fbem-1515 191 2 forests forest NOUN fbem-1515 191 3 . . PUNCT fbem-1515 192 1 machine machine NOUN fbem-1515 192 2 learning learning PROPN fbem-1515 192 3 , , PUNCT fbem-1515 192 4 45(1 45(1 NOUN fbem-1515 192 5 ) ) PUNCT fbem-1515 192 6 , , PUNCT fbem-1515 192 7 5–32 5–32 NOUN fbem-1515 192 8 . . PUNCT