Table 4.

Comparison of the k-NN classification model with other models

Model	Our study	Su et al. [42]	Wang et al. [43]	Su et al. [18]	Li et al. [44]
Method	k-NN	SVM	Naive Bayesian classifier	PLS transformed into binary QSAR	SVM
Descriptors	2D PaDEL fingerprints	2D and 3D MOE, 4D fingerprints from MD simulation	Physico-chemical property based and geometry based descriptors, and fingerprints	2D and 3D MOE descriptors and 4D fingerprints	GRIND descriptors derived from docking
Training set
Cut-off (µM)	5	–	10	40	40
Total	172	546	719	250	495
True positives	73	188	247	–	83
True negatives	48	242	315	–	283
Sensitivity	0.78	0.90	0.89	–	0.55
Specificity	0.61	0.72	0.72	–	0.83
Q	0.70	0.79	0.78	–	0.74
F-measurea	0.74	0.76	0.76	–	0.56
G-mean	0.69	0.80	0.80	–	0.67
Test set
Cut-off (%)b	20	20	20	20	20
Total	1795	1668	1953	1668	1877
True positives	140	67	135	121	107
True negatives	851	1298	1247	963	1271
Sensitivity	0.63	0.41	0.54	0.74	0.57
Specificity	0.54	0.86	0.73	0.64	0.75
Q	0.55	0.82	0.71	0.65	0.73
F-measure	0.26	0.31	0.32	0.29	0.30
G-mean	0.59	0.60	0.63	0.69	0.66

^a2[(precision*sensitivity)/(precision + sensitivity)], ^b % hERG blockage