Abstract
Recently fuzzy association rule mining techniques have been applied to classification problems to design an accurate and compact classifier. A fuzzy association rule is a way of describing a relationship in the form of if-then statements with fuzzy sets. To define the interestingness of fuzzy association rules, several criteria such as confidence and support have often been used. Association rule mining aims to extract all rules that satisfy pre-specified thresholds of confidence and support. When building a classifier from extracted fuzzy rules, it is impossible to examine all of their possible combinations. Genetic fuzzy rule selection deals with such a problem using genetic algorithms. It efficiently searches for an accurate and compact classifier. In this paper, first we demonstrate the performance of genetic fuzzy rule selection. Next, we examine obtained classifiers by genetic fuzzy rule selection. Experimental results show that Pareto-optimal rules in terms of confidence and support are often included in the obtained classifiers. Based on these observations, we propose and examine genetic fuzzy rule selection that selects fuzzy rules from only Pareto-optimal rules. Through computational experiments, we show that genetic fuzzy rule selection with only Pareto-optimal candidate rules often has almost the same accuracy and much less computation time than the case with all candidate rules.
