MEFC: MAP-Elites-based Fuzzy Classifier Design

Recently, the importance of highly transparent Artificial Intelligence (AI) has been increasing because there are many real-world applications in which the basis and internal0020process of making a decision are required when using AI. A fuzzy classifier is a highly transparent AI that can be interpreted linguistically and can make decisions considering real-world uncertainties. Multi-objective evolutionary algorithms have been actively used in fuzzy classifier design under the name of multi-objective evolutionary fuzzy systems. However, there is a possibility that multi-objective evolutionary algorithms converge to a few locally optimal solutions prematurely. Therefore, we propose a fuzzy classifier design method based on a quality diversity algorithm, which can improve diversity as well as performance. MAPElites, one of the most representative quality diversity algorithms, searches for optimal solutions while improving diversity in a predefined feature space. We show the usefulness of the proposed method, MAP-Elites-based Fuzzy Classifier design (MEFC), by comparing it with the fuzzy classifier design method based on an evolutionary multi-objective optimization algorithm. In this study, we use MEFC, which searches for accurate models in the feature space based on interpretability measures. In addition, we propose two-stage MEFC composed of the exploration stage and the exploitation stage. We show its usefulness by comparing it with the default version of MEFC.