Fair Client Selection in Federated Learning: Enhancing Fairness in Collaborative AI Systems

Ranim Bouzamoucha, Farah Barika Ktata, Sami Zhioua

2025

Abstract

Fairness in machine learning (ML) is essential, especially in sensitive domains like healthcare and recruitment. Federated Learning (FL) preserves data privacy but poses fairness challenges due to non-IID data. This study addresses these issues by proposing a client selection strategy that improves both demographic and participation fairness while maintaining model performance. By analyzing the impact of selecting clients based on local fairness metrics, we developed a lightweight algorithm that balances fairness and accuracy through a Multi-Armed Bandit framework. This approach prioritizes equitable client participation, ensuring the global model is free of biases against any group. Our algorithm is computationally simple, making it suitable for constrained environments, and promotes exploration to include underrepresented clients. Experimental results show reduced biases and slight accuracy improvements, demonstrating the feasibility of fairness-driven FL. This work has practical implications for applications in recruitment, clinical decision-making, and other fields requiring equitable, high-performing ML models.

Paper Citation

in Harvard Style

Bouzamoucha R., Ktata F. and Zhioua S. (2025). Fair Client Selection in Federated Learning: Enhancing Fairness in Collaborative AI Systems. In Proceedings of the 14th International Conference on Data Science, Technology and Applications - Volume 1: DATA; ISBN 978-989-758-758-0, SciTePress, pages 263-273. DOI: 10.5220/0013397900003967

in Bibtex Style

@conference{data25,
author={Ranim Bouzamoucha and Farah Ktata and Sami Zhioua},
title={Fair Client Selection in Federated Learning: Enhancing Fairness in Collaborative AI Systems},
booktitle={Proceedings of the 14th International Conference on Data Science, Technology and Applications - Volume 1: DATA},
year={2025},
pages={263-273},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013397900003967},
isbn={978-989-758-758-0},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 14th International Conference on Data Science, Technology and Applications - Volume 1: DATA
TI - Fair Client Selection in Federated Learning: Enhancing Fairness in Collaborative AI Systems
SN - 978-989-758-758-0
AU - Bouzamoucha R.
AU - Ktata F.
AU - Zhioua S.
PY - 2025
SP - 263
EP - 273
DO - 10.5220/0013397900003967
PB - SciTePress