Generating Interpretable Fuzzy Systems for Classification Problems

Abstract

This paper presents a new method to generate interpretable fuzzy systems from training data to deal with classification problems. The antecedent partition uses triangular sets with 0.5 interpolations avoiding the presence of complex overlapping that happens in another method. Singleton consequents are generated form the projection of the modal values of each triangular membership function into the output space. Least square method is used to adjust the consequents. The proposed method gets a higher average classification accuracy rate than the existing methods with a reduced number of rules andparameters and without sacrificing the fuzzy system interpretability. The proposed approach is applied to two classical classification problems: Iris data and the Wisconsin Breast Cancer classification problem.