Two-Stage Bagging Pruning for Reducing the Ensemble Size and Improving the Classification Performance

Table 6

Relative improvement of the proposed bagging pruning methods and existing other three variations with respect to the traditional bagging using DT as the base classifier. The values in parentheses represent the optimized parameters ta for AP, td for DP, and ta, td for “AP+DP” and “DP+AP” with the best classification performance achieved by the corresponding pruning method.

Dataset	Brag	Nice	TB	AP	“AP+DP”	DP	“DP+AP”
Aba	NA	NA	NA	1.411	2.457,5	1.567	1.562,7
Adult	NA	NA	NA	0.201	0.201,10	0.206	0.201,6
Aus	1.15	-10.34	-3.45	2.252	3.418,4	0.0010	0.000,10
Bcw	-1.37	-12.33	-5.48	3.637	7.253,7	3.639	7.253,8
Bld	NA	NA	NA	2.027	7.073,2	8.083	8.080,3
Cmc	0.00	-2.08	-2.08	0.988	2.788,5	3.191	3.625,1
Col	NA	NA	NA	0.000	2.072,9	4.147	4.141,7
Cre	NA	NA	NA	4.045	7.112,5	3.073	8.085,9
Der	NA	NA	NA	0.003	0.005,7	7.611	7.615,1
Ger	NA	NA	NA	1.732	1.732,10	0.438	0.873,9
Gla	NA	NA	NA	0.009	1.768,8	3.534	3.535,4
Hea	NA	NA	NA	4.002	4.002,10	0.0010	0.009,4
Hep	NA	NA	NA	6.899	13.792,7	10.372	13.795,6
Ion	NA	NA	NA	0.008	9.532,3	4.683	9.532,3
kr-vs-kp	NA	NA	NA	0.006	19.157,6	0.0010	6.389,2
Mam	NA	NA	NA	1.795	3.637,3	0.443	5.908,3
Pid	NA	NA	NA	1.093	2.183,2	1.645	2.182,1
Spe	NA	NA	NA	1.949	7.858,3	9.791	9.790,1
Tel	NA	NA	NA	0.501	0.504,8	0.507	0.742,7
Veh	NA	NA	NA	0.426	7.274,2	8.622	8.623,2
Vot	NA	NA	NA	13.043	21.746,5	21.743	21.741,3
Vow	NA	NA	NA	1.032	5.261,9	3.207	3.200,7
Yea	NA	NA	NA	3.136	3.975,8	2.337	2.339,7
Spambase	1.12	1.12	1.12	0.009	0.749,7	0.748	1.102,6
Tictacto	-8.47	6.78	6.78	0.003	5.812,6	18.026	18.021,6
Wdbc	3.53	-67.06	-5.88	24.798	24.798,8	4.848	9.979,9
Wpbc	2.22	-18.89	1.11	5.756	9.607,2	1.901	9.608,5
Spect	-1.08	-5.38	-3.23	1.839	7.378,4	3.663	9.253,4

Note. “NA” means not available.