Table 1 

Comparison of RSVM and SVMRFE on DataG (with gene outliers) 

Level^{a} 
ReduceSV^{b} 
P(svdiff)^{c} 
ReduceTest^{d} 
P(testdiff)^{e} 
ImproveRec^{f} 
P(recdiff)^{g} 


800 
4.01% 
1.81E42 
7.70% 
4.72E03 
3.90% 
1.71E39 
600 
5.77% 
1.74E49 
2.50% 
4.64E01 
1.70% 
5.21E15 
500 
6.83% 
2.75E51 
4.00% 
1.62E01 
0.30% 
0.079189 
400 
8.35% 
3. 26E60 
2.80% 
3.48E01 
1.10% 
4.48E06 
300 
9.33% 
3.83E58 
7.40% 
3.65E02 
3.70% 
1.77E31 
200 
8.22% 
1.28E48 
19.20% 
6.36E09 
6.30% 
5.79E44 
150 
8.55% 
1.51E53 
19.50% 
1.16E08 
7.10% 
9.76E46 
100 
4.97% 
6.20E22 
11.90% 
1.83E04 
6.00% 
6.43E40 
90 
5.84% 
1.66E27 
13.70% 
4.20E06 
4.60% 
1.07E30 
80 
5.17% 
8.20E29 
12.40% 
4.14E06 
4.50% 
7.12E29 
70 
4.14% 
1.46E27 
8.50% 
4.77E04 
3.80% 
1.05E24 
60 
3.10% 
1.23E20 
10.20% 
3.14E05 
3.40% 
4.99E24 
50 
2.27% 
2.01E15 
10.20% 
4.11E06 
2.90% 
2.37E21 


^{a }Level: The number of features selected in each recursive step. With all of the 1000 features, there is no difference between RSVM and SVMRFE because no feature selection happened. ^{b }ReduceSV: Relative reduction in the mean number of support vectors used by RSVM comparing to that by SVMRFE, calculated as: (average #SV_{SVMRFE } average #SV_{RSVM})/(average #sv_{SVMRFE}). ^{c }P(svdiff): The pvalue of the observed difference in numbers of SVs, by paired ttest. ^{d }ReduceTest: Relative reduction in the mean test error rates of SVM models with RSVMselected features comparing to that with SVMRFE selected features, calculated as: (average TestError_{SVMRFE } average TestError_{RSVM})/(average TestError_{SVMRFE}). ^{e }P(testdiff): The pvalue of the observed difference in test error rates, by paired ttest. ^{f }ImproveRec: Relative improvement in the proportion of recovered informative genes by RSVM than that by SVMRFE, calculated as: (average #REC_{RSVM } average #REC_{SVMRFE})/(average #REC_{SVMRFE}), where #REC is the number of recovered true informative genes with the method stated in the subscript. ^{g }P(recdiff): The pvalue of the observed difference in proportion of recovered true informative genes, by paired ttest. 

Zhang et al. BMC Bioinformatics 2006 7:197 doi:10.1186/147121057197 