Recursive SVM feature selection and sample classification for mass-spectrometry and microarray data

BMC Bioinformatics

Table 8 The comparison of SVM vs. WV on Data-G

Level^a	ReduceTest^d	P(test-diff)^e	ImproveRec^f	P(rec-diff)^g
800	36.36%	1.16E-17	1.02%	2.13E-06
600	38.95%	6.74E-17	9.49%	2.14E-62
500	39.51%	8.72E-21	14.82%	3.77E-71
400	44.84%	3.86E-23	20.83%	1.68E-79
300	49.75%	6.86E-25	28.72%	3.48E-86
200	54.22%	2.02E-27	36.75%	3.70E-91
150	54.83%	9.37E-30	36.14%	2.65E-86
100	43.56%	6.63E-25	33.61%	4.42E-75
90	42.35%	1.85E-26	31.09%	4.23E-73
80	37.37%	7.35E-25	29.08%	3.79E-67
70	32.23%	1.20E-20	26.54%	9.22E-63
60	27.79%	1.16E-20	24.39%	1.24E-61
50	23.47%	8.64E-15	21.80%	1.83E-53

^a Level: The number of features selected in each recursive step.
^dReduceTest: Relative reduction in the mean test error rates of SVM comparing to that of WV, calculated as: (average TestError_WV - average TestError_R-SVM)/(average TestError_WV).
^e P(test-diff): The p-value of the observed differences in test error rates, by paired t-test.
^f ImproveRec: Relative improvement in the proportion of recovered informative genes by R-SVM comparing to that by WV, calculated as: (average #REC_R-SVM - average #REC_WV)/(average #REC_WV), where #REC represents the number of recovered true informative genes with the method stated in the subscript.
^g P(rec-diff): The p-value of the observed difference in proportion of recovered informative genes, by paired t-test.

ISSN: 1471-2105