Table 3.

Algorithm comparison using the full feature set over all versus common ADRs

Method	ADR_All				ADR_50+
	AUC	ACC	Precision	Recall	AUC	ACC	Precision	Recall
LR	0.9102	0.9486	0.4152	0.5671	0.7648	0.8023	0.5321	0.6908
NB	0.9116	0.9527	0.3537	0.6302	0.8627	0.8431	0.3929	0.7214
KNN	0.9161	0.9595	0.5300	0.5787	0.8508	0.8530	0.5633	0.6401
RF	0.9491	0.9653	0.6310	0.6250	0.9052	0.8784	0.6522	0.7057
SVM	0.9524	0.9669	0.6617	0.6306	0.9141	0.8857	0.6750	0.7227

The full feature set here refers to chemical + biological + phenotypic properties. ADR_All considers all ADRs, and ADR_50+ are the common ADRs caused by at least 50 drugs. All AUC, ACC, Precision, and Recall are micro-averages across ADRs in the corresponding dataset.

ACC, accuracy; ADR, adverse drug reaction; AUC, area under the receiver operating characteristic curve; KNN, K-nearest neighbor; LR, logistic regression; NB, naïve Bayes; RF, random forest; SVM, support vector machine.