Deriving biased classifiers for better ROC performance

ROC analysis makes it possible to evaluate how well classiiers will perform given certain misclas-siication costs and class distributions. Given a set of classiiers, it also provides a method for constructing a hybrid classiier that optimally uses the available classiiers. Now in some cases it is possible to derive multiple classiiers from a single one, in a cheap way, and such that these classiiers focus on diierent areas of the ROC diagram, such that a hybrid classiier with better overall ROC performance can be constructed. This principle is quite generally applicable; here we describe a method to apply it to decision tree classiiers. An experimental evaluation illustrates the usefulness of the technique.