A Data-Centric Framework for Detecting and Mitigating Bias in Machine Learning Datasets: Metrics, Auditing, Rebalancing, and Validation

Detecting ML dataset bias is fundamentally a data-quality, measurement, and population-validity problem as well as a fairness problem. The framework proposed in this paper connects dataset profiling, representation analysis, label-quality auditing, distribution-shift detection, subgroup model evaluation, targeted mitigation, and independent validation in one repeatable workflow. The central principle is that bias should be measured before it is mitigated, and mitigation should be chosen according to the mechanism that generated the disparity. A dataset cannot be certified as unbiased by a single fairness metric, nor can a model be considered fair merely because its overall accuracy is high. Robust practice requires documented target populations, transparent assumptions, multiple complementary metrics, uncertainty estimates, external validation, and continuous monitoring. The proposed framework provides a practical foundation for building such evidence-driven ML pipelines across domains.

[1] N. Mehrabi, F. Morstatter, N. Saxena, K. Lerman, and A. Galstyan, “A Survey on Bias and Fairness in Machine Learning,” ACM Computing Surveys, vol. 54, pp. 1–35, 2019. [2] S. Caton and C. Haas, “Fairness in Machine Learning: A Survey,” ACM Computing Surveys, vol. 56, pp. 1–38, 2020. [3] T. P. Pagano, R. B. Loureiro, F. V. Lisboa, R. M. Peixoto, G. A. S. Guimarães, G. O. R. Cruz, M. M. Araujo, L. L. Santos, M. A. S. Cruz, E. L. Oliveira, I. Winkler, and E. S. Nascimento, “Bias and Unfairness in Machine Learning Models: A Systematic Review on Datasets, Tools, Fairness Metrics, and Identification and Mitigation Methods,” Big Data and Cognitive Computing, vol. 7, no. 15, 2023. [4] N. Shahbazi, Y. Lin, A. Asudeh, and H. V. Jagadish, “Representation Bias in Data: A Survey on Identification and Resolution Techniques,” ACM Computing Surveys, vol. 55, pp. 1–39, 2022. [5] A. Fabris, S. Messina, G. Silvello, and G. A. Susto, “Algorithmic fairness datasets: the story so far,” Data Mining and Knowledge Discovery, vol. 36, pp. 2074–2152, 2022. [6] C. Dwork, M. Hardt, T. Pitassi, O. Reingold, and R. Zemel, “Fairness through awareness,” in Proceedings of the 3rd Innovations in Theoretical Computer Science Conference, pp. 214–226, 2012. [7] M. Hardt, E. Price, and N. Srebro, “Equality of Opportunity in Supervised Learning,” in Advances in Neural Information Processing Systems, vol. 29, 2016. [8] A. Chouldechova, “Fair prediction with disparate impact: A study of bias in recidivism prediction instruments,” Big Data, vol. 5, no. 2, pp. 153–163, 2017. [9] J. Kleinberg, S. Mullainathan, and M. Raghavan, “Inherent trade-offs in the fair determination of risk scores,” in 8th Innovations in Theoretical Computer Science Conference, 2017. [10] R. Zemel, Y. Wu, K. Swersky, T. Pitassi, and C. Dwork, “Learning fair representations,” in International Conference on Machine Learning, pp. 325–333, 2013.