Fig. 6.

Erdos-Renyi Interaction effect simulation results. $r=100$ replicates, $n=100$ samples, $d=5000$ attributes, $k=500$ functional attributes, effect size $s=0.4$. Standard thresholdout (TO/red) is virtually ineffective (accuracy around 0.5); Private Evaporative Cooling (pEC/blue) yields the best accuracy and prevents overfitting; for reference, standard random forest mean OOB accuracy = 0.736 (Color version of this figure is available at Bioinformatics online.)