Fuzzy rule base generation for classification and its minimization via modified threshold accepting

Abstract

This paper addresses the application of a modified threshold accepting algorithm (MTA) for minimizing the number of rules in a fuzzy rule-based classification system, while guaranteeing high classification power. In terms of computational time required, the MTA outperforms the GA approaches, which are applied to this multi-objective combinatorial optimization problem in the literature. The number of rules used and the classification power are taken as the objectives. The original model of Ishibuchi et al. (IEEE Trans. Fuzzy Systems 3 1995, 260–270) is further modified by employing various aggregators such as the γ-operator (compensatory and), fuzzy and and a convex combinations of min and max operators in place of product and min operators. The performance of the present model is demonstrated in the case of Fisher's well-known Iris data and other data appearing in literature. Less computational time needed in all cases and better classification rate in testing phase (in leave-one-out technique) are important contributions of the present model.