Mitigating Unfairness via Evolutionary Multi-objective Ensemble Learning

In the literature of mitigating unfairness in machine learning, many fairness measures are designed to evaluate predictions learning models and also utilised guide training fair models. It has been theoretically empirically shown that there exist conflicts inconsistencies among accuracy multiple measures. Optimising one or several may sacrifice deteriorate other Two key questions should be considered, how simultaneously optimise measures, all considered more effectively. this paper, we view problem as a multi-objective considering A evolutionary framework is used metrics (including measures) Then, ensembles constructed based on order automatically balance different metrics. Empirical results eight well-known datasets demonstrate compared with state-of-the-art approaches for unfairness, our proposed algorithm can provide decision-makers better tradeoffs Furthermore, high-quality generated by construct an ensemble achieve tradeoff than methods. Our code publicly available at https://github.com/qingquan63/FairEMOL