Tradeoff between the performance of fuzzy rule-based classification systems and the number of fuzzy if-then rules

The main aim of this paper is to illustrate the tradeoff between the performance of a fuzzy rule-based classification system and its size (i.e., the number of fuzzy if-then rules) through computer simulations on commonly used data sets. In our computer simulations, we use a simple heuristic method for generating fuzzy if-then rules from training patterns, in which a pattern space is homogeneously partitioned into fuzzy subspaces by subdividing each axis into linguistic values. For clearly illustrating the tradeoff we use a genetic-algorithm-based rule selection method with two objectives: to minimize the number of fuzzy if-then rules and to maximize the classification performance. Various fuzzy rule-based classification systems with different sizes are generated by the rule selection method for each data set.