Fairness issues, current approaches, and challenges in machine learning models

Abstract

With the increasing influence of machine learning algorithms in decision-making processes, concerns about fairness have gained significant attention. This area now offers significant literature that is complex and hard to penetrate for newcomers to the domain. Thus, a mapping study of articles exploring fairness issues is a valuable tool to provide a general introduction to this field. Our paper presents a systematic approach for exploring existing literature by aligning their discoveries with predetermined inquiries and a comprehensive overview of diverse bias dimensions, encompassing training data bias, model bias, conflicting fairness concepts, and the absence of prediction transparency, as observed across several influential articles. To establish connections between fairness issues and various issue mitigation approaches, we propose a taxonomy of machine learning fairness issues and map the diverse range of approaches scholars developed to address issues. We briefly explain the responsible critical factors behind these issues in a graphical view with a discussion and also highlight the limitations of each approach analyzed in the reviewed articles. Our study leads to a discussion regarding the potential future direction in ML and AI fairness.