Multi-objective genetic local search for minimizing the number of fuzzy rules for pattern classification problems

For constructing compact fuzzy rule-based systems with high classification performance, we have already formulated a rule selection problem. Our rule selection problem has two objectives: to minimize the number of selected fuzzy if-then rules (i.e., to minimize the fuzzy rule base) and to maximize the number of correctly classified patterns (i.e., to maximize the classification performance). In this paper, we apply single-objective and multi-objective genetic local search algorithms to our rule selection problem. High performance of those hybrid algorithms is demonstrated by computer simulations on multi-dimensional pattern classification problems in comparison with genetic algorithms in our former studies. It is shown in computer simulations that local search procedures can improve the ability of genetic algorithms to search for a compact rule set with high classification performance.