A Novel Approach to Ensemble Classifiers: FsBoost-Based Subspace Method

Table 10

Comparing the FsBoost algorithm with single classifiers and boosting algorithms.

Methods	Datasets
	A		B		C		D
	Rank	Acc	Rank	Acc	Rank	Acc	Rank	Acc
AdaBoostM1	3	99.04	2	97.22	4	96.48	3	94.35
Bag	2	99.43	4	96.87	2	96.83	1	94.87
GentleBoost	4	98.96	3	97.04	3	96.78	2	94.65
LogitBoost	5	98.96	5	96.35	5	96.35	4	94.22
LPBoost	6	98.57	6	96.35	7	94.74	7	92.83
RobustBoost	12	95.74	16	87.78	17	86.35	15	87.17
RUSBoost	17	90.83	17	84.48	16	86.43	16	84.39
Subspace	15	94.20	10	94.15	13	92.02	11	90.54
TotalBoost	7	98.35	7	95.78	6	95.87	6	93.30
FsBoost.V1
Level 1—kNN ensemble	16	93.87	13	92.48	14	91.91	13	89.87
Level 1—PNN ensemble	14	94.26	15	87.83	15	89.17	17	80.13
Level 1—SVM ensemble	1	99.48	1	97.48	1	97.43	5	94.04
Level 2—ensemble	9	97.22	11	94.04	8	94.52	8	91.65
FsBoost.V2
Level 1—ensemble 1	11	95.91	8	94.83	12	93.30	12	90.39
Level 1—ensemble 2	8	98.00	9	94.17	10	93.39	10	91.09
Level 1—ensemble 3	13	95.52	14	91.09	11	93.35	14	89.57
Level 2—ensemble	10	96.43	12	93.52	9	93.78	9	91.30
kNN		92.17		91.96		90.00		88.04
PNN		93.26		91.00		89.04		82.43
SVMs		99.65		98.48		98.13		94.61

Acc: accuracy (%).