Navigating Ensemble Configurations for Algorithmic Fairness

TL;DR

This paper introduces an open-source library to systematically explore how ensemble methods combined with bias mitigators affect fairness and performance trade-offs in machine learning, providing practical guidance for practitioners.

Contribution

It presents a modular library for combining mitigators and ensembles, and offers empirical insights and a guidance diagram for optimizing fairness-performance trade-offs.

Findings

Ensemble configurations significantly influence fairness outcomes.

The guidance diagram helps practitioners select effective ensemble-mitigator combinations.

The approach is robust and reproducible across multiple datasets.

Abstract

Bias mitigators can improve algorithmic fairness in machine learning models, but their effect on fairness is often not stable across data splits. A popular approach to train more stable models is ensemble learning, but unfortunately, it is unclear how to combine ensembles with mitigators to best navigate trade-offs between fairness and predictive performance. To that end, we built an open-source library enabling the modular composition of 8 mitigators, 4 ensembles, and their corresponding hyperparameters, and we empirically explored the space of configurations on 13 datasets. We distilled our insights from this exploration in the form of a guidance diagram for practitioners that we demonstrate is robust and reproducible.