Table 3.

Performance of the CRF and RSVM predictors evaluated by the AUC measure of performance. For the training set the performance of the CRF is estimated by the OOB error and for the RSVM by the leave-one-out cross-validation. The predictive features selected by each method are listed in the respective columns. Both methods selection was applied to the initial set of convergent genes.

Method	CRF				RSVM
Data Set	# convergent genes	Selected features	train AUC	test AUC	Selected features	train AUC	test AUC
Golub ALL AML	80	8	1.000	0.857	60	1.000	0.893
ProstateCancer	79	5	0.963	0.958	79	0.938	0.917
BrCaMetastasis	103	8	0.862	0.815	103	0.818	0.679
antTNF-response	40	8	0.890	0.750	40	0.828	0.500