Diversity analysis on boosting nominal concepts

Abstract

In this paper, we investigate how the diversity of nominal classifier ensembles affects the AdaBoost performance [13]. Using 5 real data sets from the UCI Machine Learning Repository and 3 different diversity measures, we show that Statistic measure is mostly correlated with AdaBoost performance for 2-class problems. The experimental results suggest that the performance of AdaBoost depend on the nominal classifier diversity that can be used as a stopping criteria in ensemble learning. © 2012 Springer-Verlag.