Optimal Membership Function For Creating Fuzzy Classifiers Ensemble

Abstract

Recent researches have shown that ensembles with more diversity classifiers have more accuracy. Six methods for measuring diversity have been introduced in this paper. These methods for measuring diversity are disagreement measure, double-fault measure, Kohavi-Wolpert variance, measurement of inter-rater agreement, measure of difficulty and generalized diversity. Six methods of measuring diversity are applied to ensemble of fuzzy classifiers produced by bagging using ANFIS as the base classifier. For an ensemble of fuzzy classifiers, relationship between membership functions and diversity has been studied. Experimental results show that using triangular membership function lead to more diverse classifiers and ensemble with more accuracy.