Host: Japan Society for Fuzzy Theory and Intelligent Informatics (SOFT)
Fuzzy rule-based classifiers have been one of the most widely used classifiers because of their highly interpretable structures and high classification ability. Several different measures of a rule in a fuzzy rule-based classifier as the measure of the interestingness have been proposed and used. Among them are confidence, support, gain, variance, chi-squared value, entropy gain, gini, laplace, lift, conviction. It is shown that the best rules according to any of these measures are Pareto-optimal rules with respect to confidence and support maximization. In this paper, we the examine the effects of designing classifiers from Pareto-optimal and near Pareto-optimal rules on the classifier's accuracy-complexity tradeoff curve.
