Different Horses for Different Courses: Comparing Bias Mitigation Algorithms in ML

Prakhar Ganesh, Usman Gohar, Lu Cheng, Golnoosh Farnadi
Proceedings of the Algorithmic Fairness Through the Lens of Metrics and Evaluation, PMLR 279:96-118, 2025.

Abstract

With fairness concerns gaining significant attention in Machine Learning (ML), several biasmitigation techniques have been proposed, often compared against each other to find thebest method. These benchmarking efforts tend to use a common setup for evaluation underthe assumption that providing a uniform environment ensures a fair comparison. However,bias mitigation techniques are sensitive to hyperparameter choices, random seeds, featureselection, etc., meaning that comparison on just one setting can unfairly favour certainalgorithms. In this work, we show significant variance in fairness achieved by several al-gorithms and the influence of the learning pipeline on fairness scores. We highlight thatmost bias mitigation techniques can achieve comparable performance, given the freedomto perform hyperparameter optimization, suggesting that the choice of the evaluation pa-rameters—rather than the mitigation technique itself—can sometimes create the perceivedsuperiority of one method over another. We hope our work encourages future research onhow various choices in the lifecycle of developing an algorithm impact fairness, and trendsthat guide the selection of appropriate algorithms.

Cite this Paper

BibTeX
@InProceedings{pmlr-v279-ganesh25a, title = {Different Horses for Different Courses: Comparing Bias Mitigation Algorithms in ML}, author = {Ganesh, Prakhar and Gohar, Usman and Cheng, Lu and Farnadi, Golnoosh}, booktitle = {Proceedings of the Algorithmic Fairness Through the Lens of Metrics and Evaluation}, pages = {96--118}, year = {2025}, editor = {Rateike, Miriam and Dieng, Awa and Watson-Daniels, Jamelle and Fioretto, Ferdinando and Farnadi, Golnoosh}, volume = {279}, series = {Proceedings of Machine Learning Research}, month = {14 Dec}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v279/main/assets/ganesh25a/ganesh25a.pdf}, url = {https://proceedings.mlr.press/v279/ganesh25a.html}, abstract = {With fairness concerns gaining significant attention in Machine Learning (ML), several biasmitigation techniques have been proposed, often compared against each other to find thebest method. These benchmarking efforts tend to use a common setup for evaluation underthe assumption that providing a uniform environment ensures a fair comparison. However,bias mitigation techniques are sensitive to hyperparameter choices, random seeds, featureselection, etc., meaning that comparison on just one setting can unfairly favour certainalgorithms. In this work, we show significant variance in fairness achieved by several al-gorithms and the influence of the learning pipeline on fairness scores. We highlight thatmost bias mitigation techniques can achieve comparable performance, given the freedomto perform hyperparameter optimization, suggesting that the choice of the evaluation pa-rameters—rather than the mitigation technique itself—can sometimes create the perceivedsuperiority of one method over another. We hope our work encourages future research onhow various choices in the lifecycle of developing an algorithm impact fairness, and trendsthat guide the selection of appropriate algorithms.} }
Endnote
%0 Conference Paper %T Different Horses for Different Courses: Comparing Bias Mitigation Algorithms in ML %A Prakhar Ganesh %A Usman Gohar %A Lu Cheng %A Golnoosh Farnadi %B Proceedings of the Algorithmic Fairness Through the Lens of Metrics and Evaluation %C Proceedings of Machine Learning Research %D 2025 %E Miriam Rateike %E Awa Dieng %E Jamelle Watson-Daniels %E Ferdinando Fioretto %E Golnoosh Farnadi %F pmlr-v279-ganesh25a %I PMLR %P 96--118 %U https://proceedings.mlr.press/v279/ganesh25a.html %V 279 %X With fairness concerns gaining significant attention in Machine Learning (ML), several biasmitigation techniques have been proposed, often compared against each other to find thebest method. These benchmarking efforts tend to use a common setup for evaluation underthe assumption that providing a uniform environment ensures a fair comparison. However,bias mitigation techniques are sensitive to hyperparameter choices, random seeds, featureselection, etc., meaning that comparison on just one setting can unfairly favour certainalgorithms. In this work, we show significant variance in fairness achieved by several al-gorithms and the influence of the learning pipeline on fairness scores. We highlight thatmost bias mitigation techniques can achieve comparable performance, given the freedomto perform hyperparameter optimization, suggesting that the choice of the evaluation pa-rameters—rather than the mitigation technique itself—can sometimes create the perceivedsuperiority of one method over another. We hope our work encourages future research onhow various choices in the lifecycle of developing an algorithm impact fairness, and trendsthat guide the selection of appropriate algorithms.
APA
Ganesh, P., Gohar, U., Cheng, L. & Farnadi, G.. (2025). Different Horses for Different Courses: Comparing Bias Mitigation Algorithms in ML. Proceedings of the Algorithmic Fairness Through the Lens of Metrics and Evaluation, in Proceedings of Machine Learning Research 279:96-118 Available from https://proceedings.mlr.press/v279/ganesh25a.html.

Related Material