An overview of membership function generation techniques for pattern recognition

Abstract

The estimation of membership functions from data is an important step in many applications of fuzzy theory. In this paper, we provide a general overview of several methods for generating membership functions for fuzzy pattern recognition applications. We discuss methods based on heuristics, probability to possibility transformations, histograms, nearest neighbor techniques, feed-forward neural networks, clustering, and mixture decomposition. We also illustrate these membership generation methods using synthetic and real data sets, and discuss the suitability and applicability of these membership function generation techniques to particular situations.