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

Table 3

Comparison of the classification performance on 28 datasets between the single classifier GNB, bagging using GNB 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	GNB	Bagging	AP	“AP+DP”	DP	“DP+AP”
Aba	51.59	51.81	52.219	52.559,1	51.907	51.902,7
Adult	81.52	81.51	81.769	81.939,1	81.522	81.899,1
Aus	79.57	79.57	80.588	80.588,10	79.579	79.579,8
Bcw	93.56	93.56	93.568	93.859,1	93.564	93.719,1
Bld	54.78	54.78	61.169	62.039,1	55.945	61.459,1
Cmc	48.54	48.20	50.249	50.519,1	48.746	48.888,2
Col	36.68	36.68	38.869	45.659,1	37.232	66.308,1
Cre	80.43	80.87	81.889	82.329,5	81.012	82.329,6
Der	88.25	88.80	91.809	95.369,1	89.071	89.075,1
Ger	72.70	72.00	73.808	73.807,8	72.209	74.008,9
Gla	39.72	39.72	51.409	53.749,8	41.127	41.125,7
Hea	84.07	83.33	84.449	85.198,2	84.442	85.195,2
Hep	58.71	58.71	67.749	69.039,1	58.712	70.979,1
Ion	89.46	89.46	90.039	90.038,2	90.038	90.319,4
kr-vs-kp	62.58	62.67	64.998	64.998,6	62.862	63.679,1
Mam	78.67	78.56	79.299	79.406,6	78.774	79.296,6
Pid	75.03	75.42	75.819	76.599,4	75.559	76.209,5
Spe	69.66	71.54	72.289	73.039,1	71.913	74.919,1
Tel	72.66	72.64	72.779	72.879,3	72.677	72.869,2
Veh	43.85	44.56	46.577	46.819,2	44.5610	44.927,6
Vot	94.48	94.48	94.719	94.719,4	94.481	94.718,1
Vow	67.68	68.18	69.906	70.109,9	68.189	68.180,9
Yea	14.42	17.12	44.009	44.349,8	19.071	34.303,1
Spambase	81.72	81.70	82.039	82.489,1	81.723	82.429,3
Tictacto	69.62	69.62	70.679	70.679,10	69.627	69.733,8
Wdbc	93.85	93.85	93.855	94.386,1	94.022	94.209,1
Wpbc	62.12	64.14	67.179	71.729,1	64.146,10	76.268,1
Spect	55.43	58.43	63.679	67.049,3	60.302,10	62.179,3