1. Introduction
In training a classifier to perform a particular classification task, it is desirable to come up with a classifier that performs well on the task by achieving a low classification error rate. A common approach to obtaining such a classifier is to examine multiple classification algorithms and/or multiple configurations of one classification algorithm and then to select the single classification algorithm and configuration with the best performance. The method of cross validation is typically used to estimate the generalization error of the different classifiers and to perform the classifier selection.