Skip to main content
. 2018 Feb 2;4(2):eaao1659. doi: 10.1126/sciadv.aao1659

Fig. 5. NP-ROC bands and ROC-CV curves (generated by fivefold CV) of three classification methods in real data application 1.

Fig. 5

(A) NP-ROC bands of RFs versus penLR versus SVMs. RF dominates the other two methods for a wide range of α values. (B) ROC-CV curves of RF, penLR, and SVM. Among the three methods, RF has the largest area under the curve. (C) NP-ROC band and ROC-CV curve of RF. The black dashed vertical line marks α = 0.21, the smallest α such that the conditional type II error is no greater than 0.4. The green dashed vertical line marks α = 0.249, the value that maximizes the vertical distance from the lower curve to the diagonal line, a criterion motivated by the Youden index (39). (D) NP-ROC band and ROC-CV curve of SVM.