Fig. 5.

Main effect simulation results. $r=100$ replicate simulations, $n=100$ samples, $d=5000$ attributes, $k=500$ functional attributes, effect size $b=0.4$. Standard thresholdout (TO/red) overfits the independent validation data set; private Evaporative Cooling (pEC/blue) and thresholdout random forest (rfTO/green) give holdout accuracies that are very close to validation accuracies; for reference, standard random forest mean OOB accuracy = 0.832 (Color version of this figure is available at Bioinformatics online.)