A method for estimating the generalization ability of a model by testing it on multiple subsets of the training data.