A Maximizing-Discriminability-Based Self-Organizing Fuzzy Network for Classification Problems

A maximizing-discriminability-based self-organizing fuzzy network (MDSOFN) that can classify highly confusable patterns is proposed in this paper. The underlying notion of the proposed MDSOFN is to split the generation of fuzzy rules into linear discriminant analysis (LDA) and Gaussian mixture model (GMM). In LDA, the weights are updated by seeking directions that are efficient for discrimination. In GMM, parameter learning adopts the gradient-descent method to reduce the cost function. Since LDA-derived fuzzy rules increase the discriminative capability among different classes, the proposed MDSOFN can classify highly confusable patterns. The effectiveness of the proposed MDSOFN is demonstrated by two classification problems. A detailed comparative performance analysis for the fuzzy networks using LDA, principal component analysis (PCA), and the support vector machine (SVM), with various noise types, is presented. Experimental results and theoretical analysis indicate that the LDA-derived fuzzy network performs better than the PCA-based fuzzy network and the SVM-based fuzzy network.