Table 3.

Performance of generated QSAR classification models built using the enriched set. Q, accuracy; Qb, balanced accuracy; Sp, specificity; Se, sensitivity; and MCC, Matthew coefficient correlation.

10-Fold Cross-Validation (Full Training Set, n = 7064)
	Q	Qb	Sp	Se	MCC
CART	0.921	0.887	0.822	0.951	0.779
NN	0.890	0.826	0.707	0.947	0.684
DNN	0.941	0.917	0.962	0.873	0.836
SVM-linear	0.945	0.924	0.885	0.963	0.847
SVM-radial	0.953	0.932	0.893	0.972	0.870
SVM-sigmoid	0.939	0.916	0.873	0.959	0.831
RF	0.951	0.925	0.875	0.975	0.863
LDA	0.938	0.911	0.861	0.962	0.828
Fitting (training set, n = 7064)
	Q	Qb	Sp	Se	MCC
CART	0.930	0.900	0.845	0.956	0.805
NN	0.935	0.922	0.897	0.947	0.826
DNN	0.930	0.984	0.994	0.974	0.971
SVM-linear	0.952	0.932	0.895	0.970	0.867
SVM-radial	0.981	0.970	0.951	0.990	0.946
SVM-sigmoid	0.941	0.920	0.881	0.960	0.838
RF	0.999	0.998	0.996	0.999	0.997
LDA	0.939	0.913	0.863	0.963	0.831
External validation (test set, n = 1247)
	Q	Qb	Sp	Se	MCC
CART	0.929	0.904	0.858	0.951	0.804
NN	0.929	0.917	0.895	0.939	0.810
DNN	0.933	0.909	0.956	0.861	0.816
SVM-linear	0.949	0.926	0.882	0.970	0.857
SVM-radial	0.958	0.937	0.895	0.978	0.884
SVM-sigmoid	0.942	0.924	0.889	0.959	0.842
RF	0.949	0.926	0.882	0.970	0.857
LDA	0.937	0.906	0.848	0.964	0.823