A method of generating fuzzy classification rules with ellipsoidal regions

Abstract

This paper introduces a method of generating fuzzy classification rules from training samples. This method can decide the numbers of rules, position and shape of membership function. First, the fuzzy rule base with ellipsoidal regions is introduced. Then, the dynamic clustering arithmetic, which can dynamically separate the training samples into different clusters, is introduced. For each cluster, a fuzzy rule around a cluster center is defined. The initial tuning of rules is used by the strategy of inserting rules and aggregating rules, then the rules are tuned by genetic algorithms. This method is evaluated by two typical data sets. The accuracy of classifier by this method is comparable to the maximum accuracy of the multilayered neural network classifier, and the training time is much shorter.