Adaptive Sensitive Reweighting to Mitigate Bias in Fairness-aware Classification

TLDR

Machine learning bias and fairness have become critical concerns as data‑driven decision making spreads across sectors, yet attempts to repair training data have yielded limited success. The authors propose an iterative adaptive reweighting process that adjusts training sample weights to mitigate bias without relying on data repair. The method uses a theoretically grounded model that iteratively adapts weights for sensitive groups, addressing multiple bias types and balancing accuracy with disparate impact or mistreatment elimination. Experiments on real‑world and synthetic datasets demonstrate that the approach achieves equal or superior trade‑offs between accuracy and unfairness mitigation compared to prior fairness‑aware methods.

Abstract

Machine learning bias and fairness have recently emerged as key issues due to the pervasive deployment of data-driven decision making in a variety of sectors and services. It has often been argued that unfair classifications can be attributed to bias in training data, but previous attempts to 'repair' training data have led to limited success. To circumvent shortcomings prevalent in data repairing approaches, such as those that weight training samples of the sensitive group (e.g. gender, race, financial status) based on their misclassification error, we present a process that iteratively adapts training sample weights with a theoretically grounded model. This model addresses different kinds of bias to better achieve fairness objectives, such as trade-offs between accuracy and disparate impact elimination or disparate mistreatment elimination. We show that, compared to previous fairness-aware approaches, our methodology achieves better or similar trades-offs between accuracy and unfairness mitigation on real-world and synthetic datasets.