Fig. 2.

Comparison of classifier with previously published classifiers. The receiver operator characteristic (ROC) curves for our classifier and each previously published classifier show the sensitivity and specificity relationships for different thresholds. The ROC curve for our classifier results from a representative 10-fold cross validation repeat, matching Fig. 1. The ROC curves for the other classifiers result from leave-one-sample-out cross validation. The ROC curve for our classifier is the only one that demonstrates a strong consistent bias to the upper left, consistent with it being the only one with a statistically significant AUROC. Because Julia_8 only produces classifier scores of 0 or 1, many samples have the same score which leads to the non-stepwise behavior of the corresponding ROC curve. The dashed grey line indicates the null hypotheses of random stratification, corresponding to an AUROC of 50 %