Bias and Fairness in Ensemble Learning: A Review

Abstract

Fairness and Bias have become important topics related to Machine Learning (ML) as well as the impact they will have on the Performance of an AI System in the real world.  This article reviews the available Ensemble Learning Techniques and discusses Fairness-Aware Ensemble Learning by reviewing three general Interventions (Pre-Processing, In-Processing, Post-Processing).  Furthermore, we will look at the most popular Fairness Metrics, such as Demographic Parity, Equal Opportunity, and Equal Probability, and the role of sensitive features when evaluating potential Algorithmic Discrimination and illustrate that Fairness-Aware Ensemble Learning provides managers with significant opportunities to address Bias; however, there are still several challenges faced in achieving a balance between Prediction Accuracy and Fairness, as well as a Lack of Interpretability and Multiple Evaluation Metrics available.  Our future directions will examine Multi-Objective Frameworks that incorporate Fairness, Accuracy, and Transparency into the design of Reliable/Fair Ensemble Learning Systems.