TABLE 3

Summary of results using BestFirst descriptors in training and test sets

Set	Method	Predicted Compounds						Correctly Predicted	MCC	5-Fold CV
Set	Method			TP	FP	TN	FN	Correctly Predicted	MCC	5-Fold CV
		%
Training set	SVM^D	74	26	79	21	77	0.54
Training set	SVM^E	79	21	84	16	82	0.63
Training set	RF	100	0	100	0	100	1.00
Training set	kNN	86	14	80	20	83	0.66
Training set	C4.5/J48	98	02	97	02	97	0.95
Test set	SVM^D	73	27	73	27	73	0.46
Test set	SVM^E	73	27	78	22	75	0.51
Test set	RF	78	22	74	26	76	0.52
Test set	kNN	79	21	68	32	74	0.47
Test set	C4.5/J48	71	29	70	30	71	0.41

TP, number of true positives; FP, number of false positives; TN, number of true negatives; FN, number of false negatives; MCC, Matthews correlation coefficient; CV, cross-validation; SVM^D, support vector machine, linear model; SVM^E, support vector machine, nonlinear model; RF, random forest method; kNN, kappa nearest neighbors method; C4.5/J48, decision tree method.