On the Chance Accuracies of Large Collections of Classifiers

We provide a theoretical analysis of the chance accuracies of large collections of classifiers. We show that on problems with small numbers of examples, some classifier can perform well by random chan