Stable feature selection based on the ensemble L 1 -norm support vector machine for biomarker discovery

BMC Genomics

Table 2 Classification performance test with cross-validation

Classifier	Performance measure	FCBF	Random forest (766)	Random forest (180)	Ensemble SVM-RFE	Our method
AdaBoost	Accuracy	0.872	0.882	0.85	0.886	0.889
	F1-score	0.737	0.749	0.647	0.738	0.735
	MCC	0.668	0.677	0.568	0.676	0.686
	AUC	0.902	0.923	0.868	0.936	0.944
Logistic regression	Accuracy	0.833	0.853	0.822	0.957	0.978
	F1-score	0.704	0.722	0.664	0.915	0.958
	MCC	0.609	0.636	0.566	0.894	0.947
	AUC	0.904	0.893	0.853	0.994	0.997
Random forest	Accuracy	0.871	0.84	0.844	0.83	0.833
	F1-score	0.614	0.553	0.625	0.459	0.45
	MCC	0.579	0.504	0.557	0.473	0.457
	AUC	0.918	0.869	0.851	0.924	0.928
SVM	Accuracy	0.879	0.854	0.84	0.95	0.968
	F1-score	0.762	0.659	0.589	0.895	0.933
	MCC	0.692	0.58	0.514	0.865	0.914
	AUC	0.915	0.885	0.871	0.992	0.996

The mean performance score of 10-fold cross validation test is calculated. The items that obtained the best score are highlighted in bold text. The numbers below the random forest classifier denote the number of genes selected for the performance test

ISSN: 1471-2164