Shapley meets Rawls: an integrated framework for measuring and explaining unfairness

TL;DR

This paper introduces an integrated framework using Shapley values to measure, explain, and address unfairness in machine learning, linking fairness and explainability.

Contribution

It demonstrates how Shapley values can be used to define and explain unfairness, extending to ESL values for robustness and efficiency.

Findings

Identifies features contributing to gender unfairness in Census data.

Uses shorter computation times than bootstrap tests.

Shows the applicability of the framework to real datasets.

Abstract

Explainability and fairness have mainly been considered separately, with recent exceptions trying the explain the sources of unfairness. This paper shows that the Shapley value can be used to both define and explain unfairness, under standard group fairness criteria. This offers an integrated framework to estimate and derive inference on unfairness as-well-as the features that contribute to it. Our framework can also be extended from Shapley values to the family of Efficient-Symmetric-Linear (ESL) values, some of which offer more robust definitions of fairness, and shorter computation times. An illustration is run on the Census Income dataset from the UCI Machine Learning Repository. Our approach shows that ``Age", ``Number of hours" and ``Marital status" generate gender unfairness, using shorter computation time than traditional Bootstrap tests.