Building Combined Classifiers.

Abstract

This chapter covers different approaches that may be taken when building an ensemble method, through studying specific examples of each approach from research conducted by the authors. A method called Negative Correlation Learning illustrates a decision level combination approach with individual classifiers trained co-operatively. The Model level combination paradigm is illustrated via a tree combination method. Finally, another variant of the decision level paradigm, with individuals trained independently instead of co-operatively, is discussed as applied to churn prediction in the telecommunications industry.