Fuzzy Rule Set Discovery Using Outcome-Accumulation-Type Evolutionary Computation

Abstract

A rule set discovery method using a directed graph structure and an outcome-accumulation-type evolutionary computation strategy, GNMiner, has been proposed. Unlike conventional evolutionary computation methods that seek a single solution, GNMiner discovers a set of rules that satisfy the interestingness measures and accumulates them in a rule pool over generations. FGNMiner is proposed as an extension of GNMiner that enables the discovery of fuzzy rules without using probabilities by extending the processing at the judgement nodes of GNMiner. In this paper, FGNMiner can discover a set of fuzzy rules that satisfy the interestingness measures in a short time. The interestingness measures of the discovered rules are uniquely calculated according to the membership functions. We evaluate the quality of the fuzzy rules discovered by FGNMiner and the classification performance of classifiers using the discovered fuzzy rule set.