Multi-criteria Decision Analysis as a tool for post-processing bias mitigation in machine learning algorithms

Autor(es)

Correa, Guilherme Lima ; Campello, Betania Silva Carneiro ; Duarte, Leonardo Tomazeli

• Tipo de documento: Artigo Científico
• Fonte: COMPUTERS & INDUSTRIAL ENGINEERING; v. 210, p. 13-pg., 2025-12-25.

Resumo

This study proposes a novel post-processing bias mitigation strategy using Multicriteria Decision Analysis (MCDA) to evaluate machine learning (ML) models based on fairness and performance metrics. As ML becomes increasingly integrated into decision-making systems, concerns have emerged about unfair treatment of protected groups defined by attributes such as race, gender, or age. Fair ML techniques aim to address this issue through methods that mitigate bias at different stages of model development. Our approach operates in the post-processing stage, without requiring any modifications to the training process, data flow, or underlying algorithm. We apply MCDA to assess seven widely-used ML algorithms using real benchmark datasets and multiple fairness and accuracy criteria. The results show that MCDA enables a transparent evaluation of tradeoffs between fairness and performance. Furthermore, our method is easy to integrate into existing workflows and supports fairer model selection without the need for retraining, offering practical value for decision-makers. (AU)

• Processo FAPESP: 24/03045-0 - Abordagens avançadas em decisões multicritério sob incertezas: perspectivas metodológicas e aplicadas
• Beneficiário: Betania Silva Carneiro Campello
• Modalidade de apoio: Auxílio à Pesquisa - Regular
• Processo FAPESP: 20/09838-0 - BI0S - Brazilian Institute of Data Science
• Beneficiário: João Marcos Travassos Romano
• Modalidade de apoio: Auxílio à Pesquisa - Programa Centros de Pesquisa Aplicada