A Linguistically Interpretable ELANFIS for Classification Problems

Abstract

In this paper, a clustering based extreme learning adaptive neuro-fuzzy inference system (CELANFIS) is proposed to improve the interpretability of the neuro-fuzzy model. Sub-clustering of input-output data is done to obtain the cluster centers which are used to obtain the membership function parameters of the CELANFIS, such that it satisfies a novel distinguishability constraint, for improving the interpretability of the network. The consequent parameters are obtained using the Moore-Penrose pseudo inverse thus resulting in faster training. Benchmark real world classification problems are used to evaluate the performance of the proposed network. Performance comparison of the proposed network with the Least Square Support Vector Machine (LS-SVM) and ELANFIS shows a satisfactory tradeoff between model accuracy and interpretability.