Measuring Impact of Diversity of Classifiers on the Accuracy of Evidential Ensemble Classifiers

Abstract

Diversity being inherent in classifiers is widely acknowledged as an important issue in constructing successful classifier ensembles. Although many statistics have been employed to measure diversity among classifiers to determine whether it correlates with ensemble performance in the literature, most these measures are incorporated and explained in the non-evidential context. In this paper, we first introduce a modelling for formulating classifier outputs as triplet mass functions and an unform notation for defining diversity measures, we then present our studies on the relationship between diversity obtained by four pairwise and non-pairwise diversity measures and accuracy of classifiers combined in different orders in the framework of belief functions. Our experimental results demonstrate that the negative correlation between the diversity and accuracy is stronger than the positive one, which is not in favor of the claim that increasing diversity could lead to reduction of generalization error of classifier ensembles.