A Comparison Study on Rule Extraction from Neural Network Ensembles, Boosted Shallow Trees, and SVMs

Table 1

Datasets used in the experiments.

Dataset	#samples	#Attr.	Attr. types	Maj. class (%)	Ref.
Australian Credit Appr.	690	14	bool., cat., int., real	55.5	UCI
Breast Cancer	683	9	int.	65.0	UCI
Breast Cancer 2	569	30	int., real	62.7	UCI
Breast Canc. (prognostic)	194	33	Real	76.3	UCI
Bupa Liver Disorders	345	6	int., real	58.0	UCI
Chess (kr-versus-kp)	3196	36	bool.	52.2	UCI
Coronary Heart Disease	884	16	bool., real	64.5	[8]
German Credit	1000	20	cat., int.	70.0	UCI
Glass (binary)	163	9	Real	53.4	UCI
Haberman	306	3	int.	73.5	UCI
Heart Disease	270	13	bool., cat., int., real	55.6	UCI
ILPD (liver)	583	10	int., real	71.5	UCI
Ionosphere	351	34	int., real	64.1	UCI
Istanbul Stock Exch.	536	8	Real	54.9	UCI
Labor	57	16	cat., int., real	64.9	UCI
Musk1	476	166	int.	56.5	UCI
Pima Indians	768	8	int., real	65.1	UCI
Promoters	106	58	cat.	50.0	UCI
Saheart	462	9	bool., int., real	65.4	KEEL
Sonar	208	60	Real	53.4	UCI
Spect. Heart	267	22	bin.	58.8	UCI
Splice junct.	3175	60	cat.	51.9	LIBSVM
Svmguide	7089	4	Real	56.4	LIBSVM
Tictactoe	958	9	cat.	65.3	UCI
Vertebral Column	310	6	Real	67.7	UCI