Construction of fuzzy classification systems using multiple fuzzy rule tables

We propose a genetic algorithm-based method for adjusting the fuzzy partition of a pattern space in fuzzy classification systems with fuzzy if-then rules. We have already developed a genetic algorithm-based fuzzy partition method. In our former method, we have to select a few attributes used in a single fuzzy rule table as input variables to avoid the explosive increase in the number of generated fuzzy if-then rules. The remaining attributes are not used in the fuzzy rule table. The aim of our genetic algorithm-based method proposed in this paper is to generate a high performance classification system with multiple fuzzy rule tables. In order to restrict the number of fuzzy if-then rules within a tractable size, only a few attributes are used in each fuzzy rule table. We show the effectiveness of the proposed method by computer simulations.