Table 2.

Performance of the three classifiers (kNN, RF, NNET) using MIE predictions, chemical descriptors, and extended fingerprints as independent variables. For each method, the average number of true positives (TP), false positives (FP), true negatives (TN) false negatives (FN) and not classified (NC) were reported. The metrics to evaluate the predictivity of the models were sensitivity (SEN), specificity (SPE), balanced accuracy (BA), Matthew’s correlation coefficient (MCC), and area under the ROC curve (AUC). Performance is the average of five-fold cross-validation results obtained over 500 iterations (100 fold-splitting procedures and five parameter combinations).

Classifier	Variable	TP	FP	TN	FN	NC	SEN	SPE	BA	MCC	AUC
K-NN	MIE predictions	30.5	11.4	19.6	7.5	0.0	0.80	0.63	0.72	0.44	0.76
	Descriptors	29.5	11.5	19.5	8.5	0.0	0.78	0.63	0.70	0.41	0.76
	Fingerprints	14.6	2.3	28.5	22.2	1.4	0.40	0.92	0.66	0.37	0.75
MLP-NNET	MIE predictions	29.4	9.4	21.6	8.6	0.0	0.77	0.70	0.74	0.47	0.78
	Descriptors	30.2	9.8	21.2	7.8	0.0	0.79	0.68	0.74	0.48	0.79
	Fingerprints	28.1	12.4	18.6	9.9	0.0	0.74	0.60	0.67	0.34	0.69
RF	MIE predictions	31.1	11.6	19.2	6.4	0.7	0.83	0.62	0.73	0.47	0.77
	Descriptors	32.9	6.4	24.4	4.9	0.4	0.87	0.79	0.83	0.66	0.91
	Fingerprints	32.9	11.9	18.8	4.8	0.5	0.87	0.61	0.74	0.51	0.80