








































 

 

 

 

Implementation of Orange Data Mining to Predict Student 

Graduation on Time at Pringsewu Muhammadiyah University 
 

Roby Novianto1*, Bambang Triraharjo2, Baskoro3 
1,2,3 Sistem dan Teknologi Informasi, Universitas Muhammadiyah Pringsewu 

1robynovianto@umpri.ac.id, 2bambangtriraharjo@umpri.ac.id, 3baskoro@umpri.ac.id 

 

 

Abstract  

Thel prolcelss olf molnitolring and elvaluating thel graduatioln olf Muhammadiyah Pringselwu Univelrsity 

(UMPRI) studelnts relally ne lelds tol bel dolne l belcausel thel studelnt graduatioln ratel is an ellelmelnt olf 

accrelditatioln asselssmelnt that is velry impolrtant folr e lach Study Prolgram. Data Mining can be l use ld tol 

classify studelnt graduatioln accuracy. This study aims to l apply thel olrangel data mining applicatioln using 

thel K-Nelarelst Nelighbolr (K-NN), De lcisioln Trelel and Naivel Bayels moldells and will theln elvaluatel thel 

accuracy olf elach olf thelsel moldells. This relselarch was colnducteld at Pringselwu Muhammadiyah 

Univelrsity in selvelral batche ls, theln studelnt data will bel analyzeld using thel olrangel data mining 

applicatioln using thel K-NN, De lcisioln Trelel and Naive l Bayels moldells. Thel data te lsting prolcelss appliels 

K-Folld Cro lss Validatioln (K=5), whilel thel e lvaluatioln moldell useld is thel Colnfusioln Matrix and ROlC. 

Thel relsults olf thel colmparisoln olf thel threlel moldells arel as folllolws, K-NN has an accuracy ratel olf 75.7%, 

Delcisioln Trelel has an accuracy ratel olf 78.1%, and Naivel Baye ls has an accuracy ratel olf 77.8%. 

Thelrelfolrel, folr classifying thel graduatioln ratel olf Muhammadiyah Univelrsity studelnts, Pringselwu 

relcolmmelnds thel De lcisioln Trelel moldell belcausel it has a belttelr lelvell olf accuracy than K-NN and Naivel 

Bayels. 

 

Kelywolrds: Graduatioln, Preldictioln, Data Mining, C4.5, Naïvel Bayels 

 

I.  Introduction 

The development of the world of education in Indonesia has had the impact of very tight 

competition. This was triggered by the increasingly advanced education in universities. One of the 

impacts of competition is producing quality graduates. The criteria for quality graduates include being 

able to complete mass learning on time. The timely learning period greatly influences the quality of 

higher education [16]. The ability of universities to produce graduates who are able to solve learning 

problems on time is a factor that influences higher education accreditation. 

This is in accordance with the National Accreditation Board for Higher Education regulations 

Number 3 of 2019 concerning Higher Education Accreditation Instruments which states that one of the 

indicators for accreditation assessment is the percentage of graduates on time for each program from a 

tertiary institution. Therefore, it is necessary to monitor the student's study period. The average study 

period for students studying at Pringsewu Muhammadiyah University is still over 4 years so it is 

necessary to try an evaluation using the student classification method using the orange application with 

three models, namely K-Nearest Neighbor (KNN), Decision Tree and Naive Bayes. Length of study is 

the period of time required for students to complete their education. The duration of student study has 

been regulated in the Ministry of Education and Culture's decree regarding the undergraduate program 

(S1) education system which has a semester credit load that must be taken between 144 and 160 credits 

with a length of study on campus of between 8 and 10 semesters or the equivalent of between 4 and 5 

years. 

Information on students' grades for each semester and student graduation information can be 

processed to create data that is useful for analyzing the accuracy of students' study progress [1]. Based 

on information obtained from Muhammadiyah Pringsewa University, in the 2015 to 2020 class year 

with an average number of graduates of 63%, data obtained that the average student study period was 

P-ISSN: 2715-2448 | E-ISSN: 2715-7199 
Vol.5 No.1 Januari 2024 
Buana Information Technology and Computer Sciences (BIT and CS) 

 

Vol. 5, No.1 | 29  

mailto:robynovianto@umpri.ac.id
mailto:bambangtriraharjo@umpri.ac.id
mailto:baskoro@umpri.ac.id


 

still over 4 years [6]. Therefore, it is necessary to try to analyze the factors that support the punctuality 

and delays in the student's study period. Previous research related to predicting student graduation and 

classification mostly used K-NN, SVM, Neural Network and Naive Bayes modes. [2],[14]. On the 

other hand, many previous studies have reviewed the results of comparative analyzes of several data 

mining models used to classify poor people [1], UMKM income classification [3], nutritional 

classification [8], as well as classifications for the agricultural industry and public health [12]. 

This research aims to classify the timeliness of the study period of Muhammadiyah Pringsewu 

University students by applying three methods, namely KNN, Naive Bayes and Decision Tree. Next, a 

comparative analysis of the three models will be carried out by applying Confusion Matrix and ROC 

analysis to ensure the level of accuracy of the three methods. This research contribution really helps the 

management of Muhammadiyah University of Pringsewu to develop strategies to minimize students 

who are not on time in completing their studies and contributes to determining the accuracy performance 

of several data mining methods, including KNN, Naive Bayes and Decision Tree. 

 
II. Methods 

1. Research Flow 

This research aims to carry out a comparative analysis of the KNN, Naive Bayes and Decision Tree 

methods used to classify graduating students at Muhammadiyah University of Pringsewu. The 

application used for the simulation is Orange Data Mining, an open source data mining application that 

has been proven to be able to help researchers analyze their data. The process stages in this research 

can be seen in Figure 1. 

Figure 1 Research flow 

 

According to Figure 1, the first step is problem identification, formulation and literature review. 

This is done first to develop research objectives and research contributions [10]. Second is the process 

of collecting data, namely compiling training data and test data as a source of data classification. Third 

is the process of designing the orange data mining widget for the student graduation classification 

process and method comparison. Fourth is the process of classifying graduating students from 

Muhammadiyah University of Pringsewu using the KNN, Decision Tree and Naive Bayes models. Fifth 

is the process of evaluating the performance of classification methods and analyzing the comparison 

results of these methods. 

 

a. K-Nearest Neighbor (K-NN) 

K-Nearest Neighbor (K-NN) is a supervised method which means it requires training information to 

classify objects that are very close. The working principle of K-NN is to find the shortest distance 

between the information to be evaluated and k neighbors in the training data [11].  The dataset is 

grouped manually according to the type of student data Pringsewu Muhammadiyah University. The 

dataset used as a reference is 35 student data for which the classification model will be tested. Next, the 

formula calculates the similarity of the dataset vector to each training dataset that has been classified. 

The K-NN theorem for calculating distance universally is as follows:  

 

 

𝑑𝑖 = √∑(𝑥1𝑗 −

𝑛

𝑖=1

𝑝𝑗) 2 

 

Problem 
identification, 

problem 
formulation and 
literature review

Dataset
Orange data 

mining design

Classification 
Process using K-

NN, Naive 
Bayes and 

Decision Tree 
methods

Evaluation of 
classification 
performance 
and method 
comparison 

results

Vol. 5, No.1 | 30  



 

 

Information: 

di = Sample distance 

xij = Knowledge sample data 

pj = Data input var ke-j 

n = Number of samples 

The stages of the process of implementing the K-NN method are as follows: 

1) Determine the parameter k (number of closest neighbors). 

2) Calculates the square of the object's Euclidean distance to the given training data. 

3) Sort result number 2 in ascending order (in order from high to low) 

4) Collecting Y categories (nearest neighbor classification based on k value) 

5) By using the nearest neighbor category with the majority, the object category can be predicted. 

 

b. Decision tree (C4.5) 

Decision Tree based on C4.5 algorithm is a commonly used classification technique to extract 

relevant relationships in data. The C4.5 algorithm is a program that creates a decision tree based on a 

labeled input data set. The advantage is that the model can be easily interpreted and implemented with 

both continuous and discrete values. The C4.5 algorithm divides training data with the help of 

information acquisition [17].  Attributes that have high frequencies are considered to separate data based 

on the information available in the dataset. When calculating the gain value, you need to know the 

entropy value, namely using the following formula:  

 

𝐸𝑛𝑡𝑟𝑜𝑝𝑦 (𝑖) = 𝑓(𝑖, 𝑗). 2𝑓 [(𝑖, 𝑗)] 

 

1) Gain value using the formula: 
 

𝑔𝑎𝑖𝑛 = − . 𝐼𝐸 (𝑖) 

 

2) To calculate the gain ratio, you need to know a new term called Split Information with the 

formula:  

𝑆𝑝𝑙𝑖𝑡𝐼𝑛𝑓𝑜𝑟𝑚𝑎𝑡𝑖𝑜𝑛 = -∑𝑐
𝑡=1

𝑆1

𝑆
𝑙𝑜𝑔2

𝑆1

𝑆
 

 

3) Next, calculate the gain ratio 

 

𝐺𝑎𝑖𝑛𝑟𝑎𝑡𝑖𝑜 (𝑆, 𝐴) =  

 

4) Repeat step 2 until all records have been split. The decision tree splitting process ends 

when: 

1) All tuples in node record m are of the same class. 

2) The attributes in the dataset are not further divided. 

3) An empty branch has no records 

 

c. Naïve Bayes 

Bayesian classification is a statistical classification that can be used to predict the probability of 

membership of a class discovered by the British scientist Thomas Bayes [4]. Naive Bayes is a 

classification algorithm that is quite simple and easy to implement so this algorithm is very effective 

when tested with the correct data set, especially if Naive Bayes is combined with function selection, so 

Naive Bayes can reduce redundancies in the data, besides that Naive Bayes shows good results when 

combined with clustering methods. Naive Bayes is proven to have high accuracy compared to support 

vector machines.  

 

 

 

Vol. 5, No.1 | 31  



Vol. 5, No.1 | 31  

 

Then X is evidence, H is hypothesis, P(H|X) is probability that hypothesis H is true, evidence of true or 

hypothesis H or Posterior  

 

 

 

So variable C explains the class, while variables F1...Fn explain the character of the instructions 

in carrying out the classification. Where this formula explains the probability that the sample enters a 

special character in class C (Posterior), namely the probability that it comes out of class C (before 

entering the sample, many priors are made), multiplied by the probability of the sample character 

appearing in class (also called likelihood), divided by the probability of the character appearing global 

examples (also called evidence). The formula above can be made simply as follows  

 

𝑃𝑜𝑠𝑡𝑒𝑟𝑖𝑜𝑟 =
𝑃𝑟𝑖𝑜𝑟 𝑥 𝑙𝑖𝑘𝑒𝑙𝑖ℎ𝑜𝑑

𝑒𝑣𝑖𝑑𝑒𝑛𝑐𝑒
  

 

For continuous data classification, the Gaussian Density formula is used: 

 

 

 

Where: P: Opportunity 

Xi: Atributke i 

xi: The value of the i attribute 

Y: Class sought 

yi: Subclass Y is sought 

μ: mean, explains the average of all attributes 

σ: Standard deviation, explained variance across attributes. 

 

2. Evaluasi Kinerja 

a. Confusion matrix 

This method only uses a matrix table as in Table 1, if the dataset only consists of two classes, one 

class is considered positive and the other negative [5]. Evaluation with the confusion matrix produces 

accuracy, precision and recall values.  

 

Table 1 Confusion Matrix 

Correct 

Clalssificaltion 

Clalssified als 

+ - 

+ True positives Fallse negaltives 

- Fallse positives True negaltives 

 

True Positive is the number of positive records that are classified as positive, false positive is the 

number of negative records that are classified as positive, false negative is the number of positive 

records that are classified as negative, true negative is the number of negative records that are classified 

as negative, 

 

𝐴𝐶𝐶 =
𝑇𝑃+𝑇𝑁

𝑇𝑃+𝑇𝑁+𝐹𝑃+𝐹𝑁
 

 

 𝑃 =
𝑇𝑃

𝑇𝑃+𝐹𝑃
 

 

𝑆𝑛 =
𝑇𝑃

𝑇𝑃 + 𝐹𝑁
 

 

𝑆𝑝 =
𝑇𝑁

𝑇𝑁 + 𝐹𝑃
 



 

Vol. 5, No.1 | 32  

 

 

b. Kurva   ROC 

The ROC curve is a graphical plot that illustrates the diagnostic ability of a binary classifier system 

as its discrimination threshold varies [2]. This method was originally developed for military radar 

receiver operators starting in 1941, giving rise to its name. The ROC curve is created by plotting the 

true positive rate (TPR) against the false positive rate (FPR) at various threshold settings. True positive 

rate is also known as sensitivity, recall, or probability of detection. The false positive rate is also known 

as the false alarm probability and can be calculated as (1 - specificity). The ROC can also be thought of 

as a plot of the power as a function of the Type I Error of the decision rule (when performance is 

calculated only from a sample of the population, it can be thought of as an estimator of this quantity). 

AUC accuracy performance can be classified into several groups, namely [7]: 

1. 0.90 – 1.00 = Excellent classification 

2. 0.80 – 0.90 = Good Classification 

3. 0.70 – 0.80 = Fair Classification 

4. 0.60 – 0.70 = Poor Classification 

5. 0.50 – 0.60 = Failure Classification 

 
III. Results and Discussions 

1. Data Mining Process 
In analyzing the performance of several classification models in the Orange tool, a comparison of 

several data mining methods was carried out to select the best method with high accuracy, in classifying 

the Pringsewu Muhammadiyah University graduation status dataset as shown in Figure 3. 

 

 
Figure 3. Widget design for student graduation status classification model 

In Figure 3, a widget is designed using a classification model in Orange data mining software in the 

form of K-NN, Decision Tree and Naive Bayes which is input by a dataset that has been previously 

processed. Then the dataset is processed into classification mode. 

2. Classification Model Testing Process 

In the process of testing the classification model that has been created previously, a collection of 

test data is needed to determine the classification results as shown in Figure 4. 



Vol. 5, No.1 | 33  

 

 
 

Figure 4. Widget design for student graduation status dataset classification model 

In Figure 4 is a widget design that has been added to the classification testing process for the 

classification model. In the red box image is a set of trial data that is entered into the classification 

process to find out the results of the graduation classification of Muhammadiyah University Pringsewu 

students. 

3. Evaluation Process of Classification Model Comparison Results 

The next process is to carry out a classification model comparison process using Test and Score 

which is needed to calculate the success rate between each classification model in Orange data mining 

as shown in Figure 5. 

 

 
Figure 5. Widget design for calculating the success of the classification model 

In Figure 5 is a widget design that has been added to the process of calculating the success rate of the 

classification model using the Test and Score widget, which will then be evaluated for accuracy using 

Confusion Matrix and ROC Analysis. 

4. Simulation results of 3 classification models 

The simulation results of the classification model were carried out using a test data set with 1 

attribute as the target, 10 numeric attributes, namely NIM, Study Program, Gender, Marital Status, 

Employment Status, 3rd Sem GPA, 4th Sem GPA, 5th Sem GPA, Total SKS, Origin Students, 

Information, so that test score results are obtained as shown in Figure 6. 

 

 



 

Vol. 5, No.1 | 34  

 

 

Figure 6. Test and score widget results 

Based on student data that has been tested, the calculation results of precision, recall and accuracy for 

each model are obtained as shown in Figure 6. Model classification resultsK-NN, Decision Tree as 

well asNaive Bayes shows that the accuracy valueDecision Tree the highest is 78%. Based on Figure 

6 which also shows a comparison of 3 AUC models, it is known that the highest AUC value is the 

methodK-NN namely 0.795. AUC is used to measure discriminatory performance by estimating the 

probability of output from randomly selected examples from a positive or negative population. The 

greater the AUC, the better the classification results used. 

5. Evaluation Results with Confusion Matrix 

Confusion Matrix is a performance measurement for machine learning classification problems 

where the output can be in the form of 2 or more classes. Confusion Matrix is a table with 4 different 

mixtures of predicted values and actual values. The evaluation results for each classification model can 

be seen in Figure 7 for the K-NN model, while the Confusion Matrix results for the Decision Tree 

model can be seen in Figure 8 and the Confusion Matrix values for the Naive Bayes model can be seen 

in Figure 9. 

Figure 7. Confusion Matrix value of the K-NN method 

Figure 7 shows that the value of True Positive (TP) is 541, True Negative (TN) is 154, False Positive 

(FP) is 83, and False Negative (FN) is 140. So the Accuracy, Precision and Recall values of the K 

method -NN is as follows: 

 



Vol. 5, No.1 | 35  

 

 

 

 

 

Figure 8. Confusion Matrix value for the Decision Tree method 

Figure 8 shows that the value of True Positive (TP) is 535, True Negative (TN) is 182, False Positive 

(FP) is 89, and False Negative (FN) is 112. So the Accuracy, Precision and Recall values of the Decision 

method Tree is as follows: 

 

 

 

 

Figure 9. Confusion Matrix value of the Naive Bayes method 

Figure 9 shows that the value ofTrue Positif (TP) is 552, True Negative (TN) is 162False Positive (FP) 

is 72, andFalse Negative (FN) is 132. Then the valueAccuracy, Precision dan Recall from the 

methodNaive Bayes are as follows: 

 

 

 

 

Based on the results of evaluation and validation usingConfusion Matrix comparative values are 

obtainedAccuracy, Precision dan Recall of 3 methodsK-NN, Naive Bayes, and Decision Tree as seen 

in Table 2. 



 

Vol. 5, No.1 | 36  

 

 

Table 2. Performance Comparison 

Metode Accuracy Precision Recall 

K-NN 75,7% 74,8% 75,7% 

Nalive Balyes 77,8% 77% 77,8% 

Decision Tree 78,1% 77,7% 78,1% 

 

Based on Table 2, it can be seen that the performance of the modelNaive Bayes better than the modelK-

NN andDecision Tree. Classification accuracy cannot achieve perfect results because there must be 

error values. This is influenced by the amount of test data and training data used in the simulation 

process carried out. 

6. Evaluation Results with ROC Curve 

Manual accuracy values can be done by looking at the ROC curve comparison visualized from 

the Confusion Matrix. Model viewing ROC curves are the most easily visible way to graphically 

compare the accuracy values of each classification model. The graphical results of the ROC can be 

seen in Figures 10 and 11. Figure 10 shows that the ROC analysis results for student graduation at 

Muhammadiyah University of Pringsewu are CORRECT in each model as follows: (1)K-NN is 0.500, 

(2) Naive Bayes is 0.500, and (3) Decision Tree is 0.600. Therefore, for this case study, the model that 

has the best accuracy value isNaive Bayes andK-NN because the curve approaches the point 0.1. 

 

Figure 10. ROC analysis with the TRUE student graduation target 

Figure 11 shows that the results of the ROC analysis of LATE graduation of Pringsewu 

Muhammadiyah University students in each classification model are as follows: (1)K-NN is 

0.500, (2) Naive Bayes is 0.500, and (3) Decision Tree is 0.600. Therefore, classification 

research using 3 models with a study from Muhammadiyah University of Pringsewu is highly 

recommended using modelsNaive Bayes andK-NN because the curve approaches the point 0.1. 



Vol. 5, No.1 | 37  

 

 

Figure 11. ROC analysis with the target of LATE student graduation 

Based on the test results above, in this research the decision tree method has a slightly higher level 

of accuracy compared to the naive Bayes method. There are several analyzes that cause the decision 

tree method to have higher accuracy, including the following: 

a. Decision Trees can work well on fairly large datasets without requiring excessive computing time. 

If the dataset is large enough, Decision Tree may be more efficient then Decision Tree provides an 

easy to interpret decision tree structure, which can assist in understanding the factors that most 

influence student graduation. 

b. Naive Bayes assumes feature independence, and normality of distribution. If these assumptions are 

not fully met in the dataset, the performance of Naive Bayes can be affected. Decision Trees, in 

some cases, are more resistant to this assumption. 

Future research in the context of predicting student graduation on time using data mining methods can 

explore a number of aspects to improve the accuracy and sustainability of the model, including 

exploring the use of deep learning methods such as neural networks to understand more complex and 

non-linear patterns in the data and then considering factors time in analysis, such as changes in student 

behavior over time, curriculum changes, or changing campus policies and can develop algorithms that 

can provide better explanations for model decisions, especially in the context of academic decisions 

that can have major implications. 

 
IV. Conclusions 

The resuts of this research show that after using the K-Nearest Neighbor, Decision Tree and Naive 

Bayes models to classify the graduation status of students at Muhammadiyah University of pringsewu, 

the results obtained were that Decision Tree's performance was superior to K-Nearest Neighbor and 

Naive Bayes. It is proven that the data used by Naive Bayes has an accuracy value of 77.8%, a precision 

of 77%, while K-Nearest Neighbor has an accuracy value of 75.7%, a precision of 74% and the Decision 

Tree has an accuracy value of 77.9% and a precision of 78%. The contribution of this research can be 

used by the management of Muhammadiyah University of pringsewu to detect early the condition of 

students so that their graduation is not too late and affect the accreditation score of Muhammadiyah 

University of Pringsewu.  

 

 

 

 

 

 



 

Vol. 5, No.1 | 38  

 

References 

[1] Alim, S. (2021a). Implementasi Orange Data Mining Untuk Klasifikasi Kelulusan Mahasiswa 

Dengan Model K-Nearest Neighbor, Decision Tree Serta Naive Bayes Orange Data Mining 

Implementation For Student Graduation Classification Using K-Nearest Neighbor, Decision Tree 

And Naive Bayes Models. In Jurnal Ilmiah Nero (Vol. 6, Issue 2). 

[2] Amra, I. A. A., & Maghari, A. Y. A. (2017). Students performance prediction using KNN and 

Naïve Bayesian. ICIT 2017 - 8th International Conference on Information Technology, 

Proceedings, 909–913. doi: 10.1109/ICITECH.2017.8079967 

[3] Annur, H. (2018). Klasifikasi Masyarakat Miskin Menggunakan Metode Naïve Bayes. In Agustus 

(Vol. 10, Issue 2). 

[4] Berrar, D. (2019). Bayes’ Theorem and Naive Bayes Classifier. In S. Ranganathan, M. Gribskov, 

K. Nakai, & C. Schönbach (Eds.), Encyclopedia of Bioinformatics and Computational Biology 

(pp. 403–412). Oxford: Academic Press. doi: https://doi.org/10.1016/B978-0-12-809633-

8.20473-1 

[5] Caelen, O. (2017). A Bayesian Interpretation of the Confusion Matrix. 

[6] Eko Prasetiyo Rohmawan. (2018). Prediksi Kelulusan Mahasiswa Tepat Waktu Menggunakan 

Metode Desicion Tree Dan Artificial Neural Networ. 

[7] Forsyth, D. (2018). Probability and Statistics for Computer Science. 

[8] Hafizan, H., & Putri, A. N. (2020). Penerapan Metode Klasifikasi Decision Tree Pada Status Gizi 

Balita Di Kabupaten Simalungun (Vol. 1, Issue 2). 

[9] Kartini, D., Nugroho, R. A., & Faisal, M. R. (2017). Klasifikasi Kelulusan Mahasiswa 

Menggunakan Algoritma Learning Vector Quantization. In Jurnal Positif (Vol. 3, Issue 2). 

[10] Mikut, R., & Reischl, M. (2011). Data mining tools. WIREs Data Mining and Knowledge 

Discovery, 1(5), 431–443. doi: https://doi.org/10.1002/widm.24 

[11] Parteek Bhatia. (2019). Data Mining and Data Warehousing. 

[12] Sari Dewi. (2016). Komparasi 5 Metode Algoritma Klasifikasi Data Mining Pada Prediksi 

Keberhasilan Pemasaran Produk Layanan Perbankan. 

[13] Seref, B., & Bostanci, E. (2018). Sentiment Analysis using Naive Bayes and Complement Naive 

Bayes Classifier Algorithms on Hadoop Framework. 2018 2nd International Symposium on 

Multidisciplinary Studies and Innovative Technologies (ISMSIT), 1–7. doi: 

10.1109/ISMSIT.2018.8567243 

[14] Wati, E. F., & Rudianto, B. (2022). Universitas Bina Sarana Informatika 1 Teknik Informatika, 

Universitas Nusa Mandiri ,2 Jl. Kramat Raya No.98, Senen, Jakarta Pusat 10450 1 Jl. In Raya 

Jatiwaringin (Vol. 11, Issue 2). Teknik Dan Informatika. 

[15] Wojtek J. Krzanowski, D. J. H. (2009). ROC Curves for Continuous Data. 

[16] Kartini, K., Sujanto, B., & Mukhtar, M. (2017). The influence of organizational climate, 

transformational leadership, and work motivation on teacher job performance. IJHCM 

(International Journal of Human Capital Management), 1(01), 192-205. 

[17] Galit Shmueli, P. C. B. I. Y. N. R. P. K. C. L. Jr. (2018). Data Mining For Business Analytics. 

 


