Procedural Fairness in Machine Learning

2026-02-25 · Source: Journal of Artificial Intelligence Research · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Advanced, quick

Summary

Current machine learning fairness research primarily addresses distributive fairness, with less focus on procedural fairness. This work defines procedural fairness for ML models, drawing from philosophy and psychology, and provides formal definitions for individual and group procedural fairness. A new metric, GPFFAE (Group Procedural Fairness based on Feature Attribution Explanation), is introduced to evaluate group procedural fairness by capturing model decision processes using feature attribution explanations. The effectiveness of GPFFAE is validated across a synthetic dataset and eight real-world datasets, revealing connections between procedural and distributive fairness. Additionally, the research proposes a method to identify features causing procedural unfairness and two techniques to enhance procedural fairness, which also improve distributive fairness with minimal impact on model performance.

Key takeaway

For research scientists developing fair ML models, understanding and measuring procedural fairness is crucial. You should consider integrating GPFFAE into your model evaluation pipeline to assess not just distributive fairness, but also the fairness of the decision-making process itself. Identifying and mitigating features that contribute to procedural unfairness can lead to more robust and ethically sound models, potentially improving distributive fairness concurrently.

Key insights

Procedural fairness in ML can be formally defined and measured using feature attribution explanations.

Principles

• Procedural and distributive fairness are related.
• Feature attribution explains model decision processes.

Method

Define procedural fairness, propose GPFFAE metric using FAE, identify unfair features, and apply two methods to improve procedural fairness.

In practice

• Use GPFFAE to evaluate model fairness.
• Identify features causing unfairness.
• Apply proposed methods to enhance fairness.

Topics

• Procedural Fairness
• Machine Learning Fairness
• Feature Attribution Explanation
• Fairness Metrics
• Algorithmic Bias Mitigation

Best for: Research Scientist, AI Researcher, AI Scientist, Machine Learning Engineer

Related on AIssential

• AI Fairness Metrics Explained: A Practical Guide with Python — Measuring AI fairness requires selecting appropriate metrics as different definitions often conflict.
• Identification and Inference for Algorithmic Frontiers with Selective Labels — The paper identifies and infers fairness-accuracy frontiers despite selective label observation, using debiased machine learning.
• Statistical and Structural Approaches to Algorithmic Fairness — Algorithmic fairness must move beyond isolated individuals and deterministic audits to address systemic inequalities.
• Differential Parity: Relative Fairness Between Two Sets of Decisions — Differential parity assesses relative fairness between decision sets by measuring independence of differences from sensitive attributes.
• Demystifying the Optimal Fair Classifier in Multi-Class Classification — This work characterizes and provides algorithms to achieve the optimal accuracy-fairness frontier in multi-class classification.
• BAFIS: Dataset + Framework to assess occupational Bias and Human Preference in modern Text-to-image Models — Text-to-image models exhibit systematic occupational biases, requiring human preference feedback for fair evaluation and development.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Journal of Artificial Intelligence Research.