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

Table 4

Comparison of the classification performance on the 28 datasets between the single classifier KNN, bagging using KNN as the base learner, and the proposed corresponding pruning methods including AP, DP, “AP+DP”, and “DP+AP”. 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	KNN	Bagging	AP	“AP+DP”	DP	“DP+AP”
Aba	52.91	53.27	53.607	53.725,6	53.324	53.488,4
Adult	81.54	81.95	81.954	81.984,9	81.978	81.994,9
Aus	66.52	67.83	67.832	70.436,1	69.861	69.861,1
Bcw	94.71	94.13	94.426	94.426,3	94.288	94.565,8
Bld	61.74	63.48	64.649	65.809,8	63.4810	66.099,8
Cmc	50.10	50.51	51.054	51.261,8	50.514	50.648,5
Col	80.71	81.52	82.073	82.886,2	81.792	82.073,6
Cre	77.68	78.41	78.849	79.133,5	78.994	79.719,5
Der	96.99	97.27	97.277	97.817,1	97.272	97.275,2
Ger	67.80	68.00	68.309	68.906,2	68.002	68.807,2
Gla	73.36	71.50	75.701	75.706,7	73.366	73.365,6
Hea	77.78	78.15	79.638	80.009,5	78.896	80.379,4
Hep	83.23	81.94	81.944	83.876,6	83.236	83.879,8
Ion	82.62	82.91	84.909	84.909,10	82.911	84.339,8
kr-vs-kp	95.09	96.21	96.317	96.401,7	96.316	96.349,7
Mam	77.42	78.46	78.464	79.293,2	79.082	79.505,3
Pid	72.69	72.69	73.608	74.518,4	73.602	74.514,3
Spe	74.91	74.53	75.288	76.408,4	75.282	76.408,1
Tel	82.35	82.99	83.003	83.082,4	83.043	83.075,8
Veh	65.60	65.37	65.604	66.909,3	66.086	66.083,6
Vot	94.94	94.94	94.949	95.402,2	95.172	95.402,2
Vow	95.05	95.35	95.564	95.662,6	95.666	95.660,6
Yea	53.98	56.40	56.743	56.872,8	56.677	56.679,7
Spambase	90.00	90.74	91.027	91.229,8	90.946	91.156,3
Tictacto	82.36	84.13	84.349	84.451,3	84.137	84.457,1
Wdbc	93.32	93.32	93.320	93.504,8	93.329	93.327,1
Wpbc	70.20	70.71	72.739	74.759,2	74.241	76.268,1
Spect	78.28	79.78	80.525	82.403,1	81.271	81.659,1