Optimal Fairness under Local Differential Privacy

TL;DR

This paper develops optimal local differential privacy mechanisms to enhance fairness in data classification, establishing a theoretical link between data fairness and classification fairness, and demonstrating empirical improvements over existing methods.

Contribution

It introduces a closed-form optimal LDP mechanism for binary attributes and a framework for multi-valued attributes, linking privacy-aware pre-processing to fairness improvements.

Findings

Optimal mechanisms reduce data unfairness effectively.

Approach outperforms existing LDP fairness methods.

Maintains high accuracy close to non-private models.

Abstract

We investigate how to optimally design local differential privacy (LDP) mechanisms that reduce data unfairness and thereby improve fairness in downstream classification. We first derive a closed-form optimal mechanism for binary sensitive attributes and then develop a tractable optimization framework that yields the corresponding optimal mechanism for multi-valued attributes. As a theoretical contribution, we establish that for discrimination-accuracy optimal classifiers, reducing data unfairness necessarily leads to lower classification unfairness, thus providing a direct link between privacy-aware pre-processing and classification fairness. Empirically, we demonstrate that our approach consistently outperforms existing LDP mechanisms in reducing data unfairness across diverse datasets and fairness metrics, while maintaining accuracy close to that of non-private models. Moreover, compared with leading pre-processing and post-processing fairness methods, our mechanism achieves a more favorable accuracy-fairness trade-off while simultaneously preserving the privacy of sensitive attributes. Taken together, these results highlight LDP as a principled and effective pre-processing fairness intervention technique.