Multi-fairness Under Class-Imbalance

Arjun Roy, Vasileios Iosifidis, Eirini Ntoutsi

Abstract: Recent studies showed that datasets used in fairness-aware machine learning for multiple protected attributes (referred to as multi-discrimination hereafter) are often imbalanced. The class-imbalance problem is more severe for the protected group in the critical minority class (e.g., female +, non-white +, etc.). Still, existing methods focus only on the overall error-discrimination trade-off, ignoring the imbalance problem, and thus they amplify the prevalent bias in the minority classes. To solve the combined problem of multi-discrimination and class-imbalance we introduce a new fairness measure, Multi-Max Mistreatment (MMM), which considers both (multi-attribute) protected group and class membership of instances to measure discrimination. To solve the combined problem, we propose Multi-Fair Boosting Post Pareto (MFBPP) a boosting approach that incorporates MMM-costs in the distribution update and post-training, selects the optimal trade-off among accurate, class-balanced, and fair solutions. The experimental results show the superiority of our approach against state-of-the-art methods in producing the best balanced performance across groups and classes and the best accuracy for the protected groups in the minority class.