Comparison between Fuzzy and Interval Partitions in Evolutionary Multiobjective Design of Rule-Based Classification Systems

This paper compares fuzzy rules with interval rules through computational experiments on benchmark data sets from the UCI database using an evolutionary multiobjective rule selection method. In the design of fuzzy and interval rule-based systems for classification problems, we use three types of partitions: homogeneous fuzzy partitions, inhomogeneous entropy-based interval partitions, and inhomogeneous fuzzy partitions derived from the interval partitions. A large number of rule-based systems are designed from each type of partitions using our evolutionary multiobjective rule selection method with three objectives: to maximize the number of correctly classified training patterns, to minimize the number of rules, and to minimize the total number of antecedent conditions. Experimental results show that the fuzzification of interval rules improves their generalization ability for many data sets