Boosting Classifiers Regionally

This paper presents a new algorithm for Boosting the performance of an ensemble of classifiers. In Boosting, a series of classifiers is used to predict the class of data where later members of the series concentrate on training data that is incorrectly predicted by earlier members. To make a prediction about a new pattern, each classifier predicts the class of the pattern and these predictions are then combined. In standard Boosting, the predictions are combined by weighting the predictions by a term related to the accuracy of the classifier on the training data. This approach ignores the fact that later classifiers focus on small subsets of the patterns and thus may only be good at classifying similar patterns. In RegionBoost, this problem is addressed by weighting each classifier's predictions by a factor measuring how well that classifier performs on similar patterns. In this paper we examine several methods for determining how well a classifier performs on similar patterns. Empirical tests indicate RegionBoost produces gains in performance for some data sets and has little effect on others.