Relationships Between Accuracy and Diversity in Heterogeneous Ensemble Classifiers∗

The relationship between ensemble classifier performance and the diversity of the predictions made by ensemble base classifiers is explored in the context of heterogeneous ensemble classifiers. Specifically, numerical studies indicate that heterogeneous ensembles can be generated from base classifiers of homogeneous ensemble classifiers that are both significantly more accurate and diverse than the base classifiers. Results for experiments using several standard diversity measures on a variety of binary and multiclass classification problems are presented to illustrate the improved performance. Keywordsclassification, heterogeneous ensembles, diversity