Early antidepressant treatment response prediction in major depression using clinical and TPH2 DNA methylation features based on machine learning approaches

Table 5 Averaged prediction metrics for each classifier based on clinical characteristics and TPH2 CpGs

Feature selection	Classifier	ROC (95%CI)	F-Measure	G-Mean	Accuracy	Sensitivity	Specificity	PPV	NPV
Without RFE	Logistic	0.504 (0.482–0.525)	0.660	0.472	0.554	0.700	0.318	0.625	0.395
	Rpart	0.530 (0.518–0.543)	0.669	0.486	0.566	0.710	0.332	0.633	0.414
	SVM-RBF	0.508 (0.496–0.521)	0.759	0.094	0.613	0.985	0.009	0.617	0.270
	LogitBoost	0.570 (0.558–0.582)	0.690	0.537	0.598	0.712	0.405	0.660	0.464
	RF	0.611 (0.599–0.623)	0.744	0.431	0.625	0.880	0.211	0.644	0.520
With RFE	Logistic	0.590 (0.569–0.611)	0.707	0.446	0.589	0.799	0.249	0.633	0.433
	Rpart	0.559 (0.547–0.572)	0.708	0.506	0.605	0.775	0.331	0.653	0.476
	SVM-RBF	0.630 (0.625–0.635)	0.753	0.378	0.627	0.918	0.156	0.638	0.540
	LogitBoost	0.617 (0.612–0.623)	0.717	0.583	0.635	0.742	0.458	0.689	0.523
	RF	0.729 (0.718–0.740)	0.785	0.598	0.700	0.882	0.405	0.706	0.679

ISSN: 1471-244X