id	author	title	date	pages	extension	mime	words	sentence	flesch	summary	cache	txt
ajis02-1539	Adnan, Md Nasim; Islam, Md Zahidul	ForEx++: A New Framework for Knowledge Discovery from Decision Forests	2017	20	.pdf	application/pdf	10720	546	63	Criteria J48 rules RS Forest PA WOBS RF+HC rules ForEx++ rules RF+HC rules ForEx++ rules Total Rules 11 62 62 55 55 Average Accuracy 76.73% 95.24% 97.84% 92.28% 98.06% Average Coverage 9.09% 5.75% 7.10% 5.62% 5.50% Average Rule Length 3.36 4.52 3.98 4.62 4.27 Distinct Root Nodes 1 3 4 3 3 Distinct Class Values 2 2 3 2 3 Table 3: Comparison among J48, RF+HC and ForEx++ rules from the OASIS data set From Table 3, we see that J48 rules have more average Cov than RF+HC and ForEx++ rules and this is due to a low number of rules (11) generated by the J48 tree from OASIS. In this paper we propose a new, data set independent framework (ForEx++) for extracting a subset of decision forest rules that are comparatively more accurate, generalized and concise than others.	cache/ajis02-1539.pdf	txt/ajis02-1539.txt