Evolutionary modular fuzzy system

Generalization is one of the most important issues in designing fuzzy systems using evolutionary computational techniques. It is not always true that the evolved system with the highest fitness has the best generalization ability. Generally if is difficult if not impossible, to tell which system among the final population of evolved systems has the best generalization ability. An evolutionary modular fuzzy system is proposed. Instead of selecting a single system, a set of systems is selected from the final population. The selected systems are combined together with each serving as a module of the final system and having a contribution to the final system's performance proportional to its fitness. Preliminary simulation studies are presented to illustrate the effectiveness of this approach.