Host: Japan Society for Fuzzy Theory and Intelligent Informatics (SOFT)
Genetic fuzzy rule selection can obtain a simple and accurate fuzzy rule set with linguistically interpretable rules from numerical data. In the first phase of our method, a large number of fuzzy rules are extracted from numerical data by a data mining technique. In the second phase, a small number of the extracted rules are selected by a genetic algorithm. When we apply this method to large data, each solution needs much longer time for evaluation. To cope with this problem, we proposed the parallel distributed implementation of genetic fuzzy rule selection. In our proposed method, not only population but also training patterns are divided into subgroups. Each subgroup is assigned to one CPU. The proposed method can drastically reduce the computational time without severer deterioration of classification performance. In this paper, we examine the effects of data subdivision on the classification performance and computational time.
