Recursive gene selection based on maximum margin criterion: a comparison with SVM-RFE

BMC Bioinformatics

Table 2 Performance comparison for binary-class datasets (continued).

Classifier+Selection criterion	Number of genes
	10	20	30	50	100
Medulloblastoma
NMC+S2N	42.1 ± 1.1	40.9 ± 1.0	40.1 ± 0.9	40.8 ± 1.0	39.3 ± 1.1
NMC+MMC-RFE(U)	39.0 ± 1.0	36.5 ± 1.1	36.5 ± 1.0	35.8 ± 0.9	35.2 ± 1.0
NMC+MMC-RFE(O)	39.7 ± 0.9	37.1 ± 0.9	34.7 ± 0.9	33.2 ± 0.9	32.4 ± 0.9
NMC+SVM-RFE(H)	42.2 ± 1.1	38.5 ± 1.0	37.5 ± 1.0	34.8 ± 0.9	34.3 ± 0.9
NMC+SVM-RFE(S)	35.3 ± 0.9	32.8 ± 0.9	32.3 ± 0.9	31.5 ± 0.9	31.0 ± 0.9
MMC+MMC-RFE(U)	38.8 ± 0.9	36.9 ± 1.0	36.4 ± 1.0	35.8 ± 0.9	35.3 ± 1.0
MMC+MMC-RFE(O)	40.0 ± 0.9	37.0 ± 0.9	34.0 ± 0.9	32.9 ± 0.9	32.2 ± 0.9
SVM+SVM-RFE(H)	41.0 ± 1.0	37.9 ± 0.9	36.8 ± 0.9	35.7 ± 0.9	36.0 ± 0.9
SVM+SVM-RFE(S)	34.6 ± 0.4	32.9 ± 0.6	33.2 ± 0.8	33.9 ± 0.8	34.6 ± 0.8
Breast cancer
NMC+S2N	34.2 ± 0.8	34.5 ± 0.8	35.0 ± 0.8	35.9 ± 0.8	36.1 ± 0.8
NMC+MMC-RFE(U)	38.0 ± 0.8	37.3 ± 0.7	36.8 ± 0.8	36.7 ± 0.7	35.4 ± 0.7
NMC+MMC-RFE(O)	37.7 ± 0.7	36.4 ± 0.7	35.6 ± 0.7	34.8 ± 0.7	35.2 ± 0.7
NMC+SVM-RFE(H)	39.4 ± 0.8	37.8 ± 0.7	36.6 ± 0.8	36.5 ± 0.7	35.6 ± 0.7
NMC+SVM-RFE(S)	36.6 ± 0.9	34.4 ± 0.8	34.1 ± 0.7	33.8 ± 0.7	33.4 ± 0.7
MMC+MMC-RFE(U)	38.5 ± 0.9	39.3 ± 0.7	38.2 ± 0.7	38.4 ± 0.7	37.2 ± 0.8
MMC+MMC-RFE(O)	38.0 ± 0.8	38.2 ± 0.8	37.0 ± 0.7	38.0 ± 0.7	36.9 ± 0.7
SVM+SVM-RFE(H)	41.1 ± 1.0	41.3 ± 0.9	41.7 ± 1.0	40.8 ± 0.8	40.7 ± 0.8
SVM+SVM-RFE(S)	43.4 ± 0.3	38.2 ± 0.6	36.3 ± 0.7	34.8 ± 0.7	35.0 ± 0.7

The average error and standard error rates (%) for Medulloblastoma and Breast cancer, when the number of genes is {10, 20, 30, 50, 100}. SVM-RFE(S) shows the best result with respect to the C parameter; NMC+SVM-RFE(S): C = 0.001, SVM+SVM-RFE(S): C = 0.01 for Medulloblastoma; NMC+SVM-RFE(S): C = 0.001, SVM+SVM-RFE(S): C = 0.001 for Breast cancer.

ISSN: 1471-2105