Bayes-Optimal Fair Classification with Linear Disparity Constraints via Pre-, In-, and Post-processing

2024-02-05

arXiv.org

AI Summary

This paper addresses unfair impacts on protected groups in machine learning by proposing a Bayesian-optimal fair classification framework that minimizes overall classification error subject to linear or bilinear group fairness constraints (e.g., demographic parity, equalized odds). Methodologically, it unifies preprocessing, in-processing, and post-processing approaches within a single probabilistic framework—incorporating calibrated probability estimation, group-specific thresholding, fairness-aware cost-sensitive learning, plug-in estimators, and resampling. Its key theoretical contribution is the first formal connection between Bayesian-optimal fair classification and the Neyman–Pearson lemma, enabling rigorous treatment of multiple fairness constraints—even when sensitive attributes are unavailable at test time. Experiments across multiple benchmark datasets demonstrate that the method achieves near-Bayesian-optimal fairness–accuracy trade-offs, substantially outperforming state-of-the-art baselines while enabling precise control over various group disparity metrics.

Technology Category

• Machine Learning: Bayesian Learning
• Computer Vision: Bias, Fairness & Privacy
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Fairness, privacy, and diversity in economic environments
• Social Networks and Social Media: Fairness and bias in social network and social media analysis

Abstract

Machine learning algorithms may have disparate impacts on protected groups. To address this, we develop methods for Bayes-optimal fair classification, aiming to minimize classification error subject to given group fairness constraints. We introduce the notion of emph{linear disparity measures}, which are linear functions of a probabilistic classifier; and emph{bilinear disparity measures}, which are also linear in the group-wise regression functions. We show that several popular disparity measures -- the deviations from demographic parity, equality of opportunity, and predictive equality -- are bilinear. We find the form of Bayes-optimal fair classifiers under a single linear disparity measure, by uncovering a connection with the Neyman-Pearson lemma. For bilinear disparity measures, Bayes-optimal fair classifiers become group-wise thresholding rules. Our approach can also handle multiple fairness constraints (such as equalized odds), and the common scenario when the protected attribute cannot be used at the prediction phase. Leveraging our theoretical results, we design methods that learn fair Bayes-optimal classifiers under bilinear disparity constraints. Our methods cover three popular approaches to fairness-aware classification, via pre-processing (Fair Up- and Down-Sampling), in-processing (Fair Cost-Sensitive Classification) and post-processing (a Fair Plug-In Rule). Our methods control disparity directly while achieving near-optimal fairness-accuracy tradeoffs. We show empirically that our methods compare favorably to existing algorithms.

Problem

• Developing Bayes-optimal fair classification with fairness constraints
• Addressing disparate impacts on protected groups in machine learning
• Creating pre-, in-, and post-processing methods for fair classification

Innovation

• Bayes-optimal fair classification with linear constraints
• Group-wise thresholding rules for bilinear disparities
• Pre-, in-, and post-processing methods for fairness