LEARNING OF FUZZY CLASSIFICATION RULES BY A GENETIC ALGORITHM

For function approximation problems using fuzzy rules, Nomura and coll eagues proposed a method of fuzzy partitioning of an input space by a genetic algorithm. In this paper, we apply their method to pattern cla ssification problems. We extend the coding method in their work, which could handle only triangular fuzzy sets, to the case in which interva ls and trapezoidal fuzzy sets can be used as antecedent fuzzy sets. By this extension, a ''don't care attribute'' can be handled, and the nu mber of fuzzy rules required for the pattern classification can be red uced. To efficiently decrease the number of fuzzy rules, different mut ation probabilities are assigned to mutation operations for increasing and for decreasing the number of membership functions in our genetic algorithm. An additional mutation operation is also introduced for fin e-tuning the shape of each membership function. Moreover we propose a hybrid algorithm that simultaneously adjusts antecedent fuzzy sets and the grade of certainty of each fuzzy classification rule.