Table 5. Summary of results.

Shown are out-of-bootstrap accuracy, sensitivity, specificity, and AUC (mean and 95% CI [59]), along with .632+ error [58] based on 100 bootstrapped trials using the best performing feature-based classifiers (all with RF feature selection) as well as those of the scalar metrics from univariate logistic regression.

	Deep learning	SVM	RF	BrIC	CSDM-WB	CSDM-CC	Peak-CC
Accuracy	0.800 (0.650–0.915)	0.767 (0.614–0.903)	0.784 (0.628–0.903)	0.781 (0.636–0.950)	0.75 (0.600–0.882)	0.754 (0.583–0.882)	0.678 (0.522–0.833)
Sensitivity	0.766 (0.414–1.000)	0.713 (0.308–1.000)	0.769 (0.348–1.000)	0.665 (0.333–1.000)	0.671 (0.375–1.000)	0.734 (0.300–1.000)	0.588 (0.250–1.000)
Specificity	0.835 (0.617–1.000)	0.820 (0.554–1.000)	0.806 (0.565–1.000)	0.878 (0.667–1.000)	0.816 (0.600–1.000)	0.784 (0.500–1.000)	0.763 (0.533–1.000)
AUC-Testing	0.850 (0.729–0.979)	0.846 (0.714–0.977)	0.847 (0.712–0.986)	0.803 (0.643–0.974)	0.835 (0.664–1.000)	0.818 (0.600–0.967)	0.778 (0.568–0.926)
.632+ error	0.148	0.176	0.163	0.207	0.246	0.227	0.292