Demystifying the Optimal Fair Classifier in Multi-Class Classification

2026-05-30

AI Summary

This work addresses the challenge of achieving fairness in multiclass classification without compromising accuracy, a problem for which existing methods lack both practical solutions and theoretical characterization of the optimal fair classifier. The study presents the first characterization of the Pareto-optimal frontier between accuracy and fairness in the multiclass setting and introduces two practical algorithms that operate without access to sensitive attributes: an in-processing fairness intervention based on reduction-based optimization and a post-processing probability calibration method grounded in plug-in estimation. Both approaches approximate the optimal frontier by solving fairness-constrained probabilistic models. Theoretically, the proposed methods are shown to converge to this Pareto-optimal frontier, and extensive experiments across multiple datasets demonstrate their superior ability to balance fairness and classification performance compared to existing alternatives.

Abstract

Ensuring fair and equitable treatment across diverse groups, particularly in multi-class classification tasks, poses a significant challenge due to the persistent biases inherent in machine learning models. Most existing bias mitigation techniques are tailored to binary settings, and the presence of multi-dimensional outputs and complex fairness mechanisms makes their extension to multi-class scenarios neither straightforward nor effective. In this paper, we investigate two fundamental, unresolved challenges in fair classification: (i) characterizing the optimal accuracy-fairness frontier in multi-class settings, and (ii) designing practical algorithms that attain this optimum in different training phases. To tackle these challenges, we first specify an analytically tractable probabilistic formulation of the optimal classifier under fairness constraints. Building upon this, we propose two attribute-blind algorithms to enforce fairness requirements in practice: an in-processing approach for fairness intervention during training via the reduction approach, and a post-processing approach for fine-tuning output probabilities with plug-in estimation. Theoretical analysis reveals that both methods converge to the optimal accuracy-fairness Pareto frontier. Experiments conducted on multiple datasets demonstrate the superior performance of our methods in balancing accuracy and fairness.

Problem

• fair classification
• multi-class classification
• accuracy-fairness trade-off
• optimal fairness
• bias mitigation

Innovation

• multi-class fair classification
• accuracy-fairness trade-off
• attribute-blind algorithms
• in-processing fairness
• post-processing calibration