Models under which random forests perform badly; consequences for applications

José A. Ferreira

We give examples of data-generating models under which Breiman's random forest may be extremely slow to converge to the optimal predictor or even fail to be consistent. The evidenc…