An Empirical Study of Building Compact Ensembles

Abstract

Ensemble methods can achieve excellent performance relying on member classifiers’ accuracy and diversity. We conduct an empirical study of the relationship of ensemble sizes with ensemble accuracy and diversity, respectively. Experiments with benchmark data sets show that it is feasible to keep a small ensemble while maintaining accuracy and diversity similar to those of a full ensemble. We propose a heuristic method that can effectively select member classifiers to form a compact ensemble. The idea of compact ensembles is motivated to use them for effective active learning in tasks of classification of unlabeled data.