Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FALE: Fairness-Aware ALE Plots for Auditing Bias in Subgroups

Giorgos Giannopoulos, Dimitris Sacharidis, Nikolas Theologitis, Loukas Kavouras, Ioannis Emiris

TL;DR

Subgroup fairness remains challenging to audit and explain. The paper introduces FALE plots, which extend accumulated local effects to measure how feature values influence subgroup unfairness under a selected fairness notion and sensitive attribute. FALE(x_i) aggregates fairness differences across bins to produce a visual, population-informed summary of bias shifts. The authors demonstrate FALE on the Adult dataset with an XGBoost classifier, showing how bias against the sex attribute varies across age, education, and work hours, and argue that FALE provides a practical first-step tool for practitioners to identify problematic subgroups.

Abstract

Fairness is steadily becoming a crucial requirement of Machine Learning (ML) systems. A particularly important notion is subgroup fairness, i.e., fairness in subgroups of individuals that are defined by more than one attributes. Identifying bias in subgroups can become both computationally challenging, as well as problematic with respect to comprehensibility and intuitiveness of the finding to end users. In this work we focus on the latter aspects; we propose an explainability method tailored to identifying potential bias in subgroups and visualizing the findings in a user friendly manner to end users. In particular, we extend the ALE plots explainability method, proposing FALE (Fairness aware Accumulated Local Effects) plots, a method for measuring the change in fairness for an affected population corresponding to different values of a feature (attribute). We envision FALE to function as an efficient, user friendly, comprehensible and reliable first-stage tool for identifying subgroups with potential bias issues.

FALE: Fairness-Aware ALE Plots for Auditing Bias in Subgroups

TL;DR

Subgroup fairness remains challenging to audit and explain. The paper introduces FALE plots, which extend accumulated local effects to measure how feature values influence subgroup unfairness under a selected fairness notion and sensitive attribute. FALE(x_i) aggregates fairness differences across bins to produce a visual, population-informed summary of bias shifts. The authors demonstrate FALE on the Adult dataset with an XGBoost classifier, showing how bias against the sex attribute varies across age, education, and work hours, and argue that FALE provides a practical first-step tool for practitioners to identify problematic subgroups.

Abstract

Fairness is steadily becoming a crucial requirement of Machine Learning (ML) systems. A particularly important notion is subgroup fairness, i.e., fairness in subgroups of individuals that are defined by more than one attributes. Identifying bias in subgroups can become both computationally challenging, as well as problematic with respect to comprehensibility and intuitiveness of the finding to end users. In this work we focus on the latter aspects; we propose an explainability method tailored to identifying potential bias in subgroups and visualizing the findings in a user friendly manner to end users. In particular, we extend the ALE plots explainability method, proposing FALE (Fairness aware Accumulated Local Effects) plots, a method for measuring the change in fairness for an affected population corresponding to different values of a feature (attribute). We envision FALE to function as an efficient, user friendly, comprehensible and reliable first-stage tool for identifying subgroups with potential bias issues.
Paper Structure (9 sections, 1 equation, 1 figure)

This paper contains 9 sections, 1 equation, 1 figure.

Table of Contents

  1. Introduction
  2. Related Work
  3. Fairness-Aware ALE Plots
  4. Preliminaries.
  5. Fairness.
  6. Fairness-Aware ALE.
  7. Demonstration
  8. Conclusion
  9. Acknowledgements

Figures (1)

  • Figure 1: FALE plots for the fairness definition of statistical parity on the sensitive attribute sex in the adult dataset. Each FALE plot examines the influence of a different attribute. Fig. 1 (a) depicts the contribution of the age attribute, Fig. 1 (b), depicts the contribution of the education attribute and, Fig. 1 (c) illustrates the influnce of the hours per week attribute. Positive FALE effects indicate additional bias from the model towards females, whereas negative FALE effects indicate less bias towards females compared to the average population.