Effects of Accuracy-based Single-Objective Optimization in Multiobjective Fuzzy Genetics-based Machine Learning

Fuzzy classifier design often requires the maximization of classification accuracy and minimization of model complexity. Multiobjective Fuzzy Genetics-Based Machine Learning (MoFGBML) can efficiently obtain a set of fuzzy classifiers considering the above-mentioned two objectives simultaneously using an evolutionary multiobjective optimization algorithm. However, the search by MoFGBML is biased toward minimizing the complexity, and it is easy to obtain classifiers with low complexity. At the same time, it is difficult to obtain classifiers with high classification accuracy. In this paper, we propose a two-stage MoFGBML, which first performs accuracy-oriented single-objective optimization to obtain a set of accurate classifiers with a large number of rules. Then, multiobjective optimization is performed to obtain a wide variety of classifiers, from highly-accurate ones to very simple ones.