Wasserstein-based fairness interpretability framework for machine learning models

TL;DR

This paper introduces a Wasserstein-based interpretability framework to measure and explain bias in machine learning models at the distribution level, utilizing transport theory and cooperative game theory for detailed bias analysis.

Contribution

It presents a novel framework combining Wasserstein metrics and transport theory to quantify and decompose model bias, incorporating fairness considerations and additivity through game theory techniques.

Findings

Effective bias quantification across sub-populations

Decomposition of bias into positive and negative contributions

Enhanced interpretability of model fairness

Abstract

The objective of this article is to introduce a fairness interpretability framework for measuring and explaining the bias in classification and regression models at the level of a distribution. In our work, we measure the model bias across sub-population distributions in the model output using the Wasserstein metric. To properly quantify the contributions of predictors, we take into account the favorability of both the model and predictors with respect to the non-protected class. The quantification is accomplished by the use of transport theory, which gives rise to the decomposition of the model bias and bias explanations to positive and negative contributions. To gain more insight into the role of favorability and allow for additivity of bias explanations, we adapt techniques from cooperative game theory.