Heuristic Rule Weight Specification for Fuzzy Rule-Based Classification Systems

Abstract

In This paper, we examine the performance of heuristic methods for rule weight specification in fuzzy rule-based classification systems. First we describe two existing rule weight specification methods using the terminology in data mining. We explain fuzzy versions of two measures of association rules in data mining: confidence and support. Next we propose two heuristic methods for rule weight specification. Diffierences among the four heuristic methods are visually demonstrated through computer simulations on an artificial test problem. Then we examine the classification performance of fuzzy rule-based systems designed by each weight specification method through computer simulations on six real world data sets. Simulation results show that the proposed two heuristic methods clearly outperform the existing ones. Finally we show that a small number of simple fuzzy rules with rule weights, which are selected by a genetic algorithm, have high interpretability and high classification ability.