Fairness-aware Classifier Design via Multi-objective Fuzzy Genetics-based Machine Learning

Takeru Konishi, Naoki Masuyama, Jorge Casillas, Yusuke Nojima

Abstract: There is a growing interest in fairness and transparency in Artificial Intelligence (AI) in order to use AI with confidence. In addition, the consideration of fairness in Explainable AI (XAI) has attracted much attention. A fuzzy classifier is one of the most representative XAI methods that can linguistically explain the basis of classification. Multi-objective Fuzzy Genetics-based Machine Learning (MoFGBML) can efficiently obtain a set of fuzzy classifiers considering the maximization of classification accuracy and minimization of model complexity simultaneously by Evolutionary Multi-objective Optimization Algorithms (EMOAs). In this study, we investigate the effect of the conventional MoFGBML on the fairness of the obtained fuzzy classifiers and examine the effect of fairness-aware optimization on the search performance of MoFGBML. Analysis of the results provided many insights into fairness-aware MoFGBML.