Fuzzy classifiers tuning using genetic algorithms with FCM-based initialization

Abstract

This paper presents an initialization technique for a Simple Genetic Algorithm that tunes a Fuzzy Inference System working as a classifier. The proposed technique uses the Fuzzy C-Means (FCM) clustering algorithm to generate the initial population of a Simple Genetic Algorithm. Two classification problems are considered to validate the proposed algorithm and to compare it against a Simple Genetic Algorithm with random initialization. Results show that it is possible to achieve a reduction in generations necessary for finding a desired classifier by using the proposed technique.