Genetic-algorithm-based approaches to the design of fuzzy systems for multi-dimensional pattern classification problems

In this paper, we examine two genetic-algorithm based approaches to the design of fuzzy-rule-based systems for multi-dimensional pattern classification problems. One approach handles a set of fuzzy if-then rules as an individual in genetic algorithms. A fitness value is assigned to each rule set, and a crossover operator is applied to a pair of rule sets. The other approach is a fuzzy classifier system where a single fuzzy if-then rule is handled as an individual. A fitness value is assigned to each fuzzy if-then rule, and a crossover operator is applied to a pair of rules. The main aim of this paper is to examine the ability of these two approaches to design a fuzzy-rule-based system with high classification performance. This examination is done by computer simulations on a real-life pattern classification problem. Moreover the classification performance of fuzzy-rule-based systems is compared with that of non-fuzzy classification methods.