Rule Evaluation and Remove Rule Selection in Michigan-style Fuzzy Genetics-based Machine Learning

Abstract

Michigan-type genetics-based machine learning has been actively studied as one of the most well-known classifier design approaches. Its fuzzy version has also been studied for data sets with continuous attributes. In this approach, each individual in the population represents a rule, while the population is regarded as a classifier. At every generation, a number of offspring rules are generated and replace unsuitable rules in the population. In this paper, we compare several methods for selection of these unsuitable rules with each other through computational experiments.