PSO with mutation for fuzzy classifier design

Abstract

Abstract One of the important issues in the design of fuzzy classifiers is the formation of fuzzy if-then rules and the membership functions. This paper presents a hybrid Particle Swarm Optimization based approach for fuzzy classifier design which incorporates the concept of mutation from evolutionary computations. The proposed MutPSO develops the fuzzy classifier system by encoding and evolving both the membership functions and rule set as particles simultaneously. Non-uniform mutation is applied to the membership functions which are represented as real numbers. Uniform mutation is applied to the rule set which is represented as discrete numbers. In the classification problem under consideration, the objective is to maximize the correctly classified data and minimize the number of rules. This objective is formulated as a fitness function to guide the search procedure to select an appropriate fuzzy classification system so that the number of fuzzy rules and the number of incorrectly classified patterns are simultaneously minimized. The performance of the proposed MutPSO approach is demonstrated through development of fuzzy classifiers for iris data available in UCI machine learning repository. Simulation results show the suitability of the proposed approach for developing the fuzzy system.