Mitigating Algorithmic Gender Bias in Brazilian Tech Salary Prediction

• Ana Luiza Bocato da Silva UNIFESP
• Lilian Berton UNIFESP

DOI: https://doi.org/10.5753/eniac.2026.31050

Resumo

This study investigates algorithmic gender bias in salary forecasting within the Brazilian tech sector using the 2024 State of Data Brazil survey. Baseline Decision Tree and Logistic Regression models were trained to evaluate how historical wage disparities are structurally replicated. An accuracy paradox emerged: high overall female accuracy (75.10%) masked a critical recall of only 39.44% for high-earning women. This low recall nearly matches the male false-positive rate (29.21%), easily overestimating lower-earning male professionals. The “omission hypothesis” was also tested, showing that selective demographic non-disclosure by high-earning men skews training data. To fix these distortions, an in-processing algorithmic mitigation framework was implemented via the Fairlearn toolkit. The adjustment drastically reduced the Demographic Parity difference (0.29 to 0.050) and Equal Opportunity difference (0.42 to 0.043). Recall for high-earning female professionals surged significantly from 39.44% to 77.46%. The equity gains incurred a negligible global accuracy loss of less than 1% (dropping from 74.90% to 73.95%). Findings prove proactive constraints eliminate automated gender penalties in labor analytics without compromising model utility.

Referências

Agarwal, A., Beygelzimer, A., Dudík, M., Langford, J., and Wallach, H. (2018). A reductions approach to fair classification. In International Conference on Machine Learning, pages 60–69. PMLR.

Barocas, S., Hardt, M., and Narayanan, A. (2019). Fairness and Machine Learning: Limitations and Opportunities. mitpress.ubiquity.press.

Bird, S., Dudík, M., Edgar, R., Errouimi, B., Khanna, R., Masarek, D., Dudík, M., Vaughan, J. W., Madaio, M., and Others (2020). Fairlearn: A toolkit for assessing and improving fairness in ai. In Microsoft Technical Report, volume MSR-TR-2020-32.

Caton, S. and Haas, C. (2020). Fairness in machine learning: A survey. arXiv preprint arXiv:2010.04053.

Data Hackers (2024). State of data brazil 2024/2025. [link]. Accessed: June 2026.

Datta, A., Tschantz, M. C., and Datta, A. (2015). Proxy non-discrimination in data-driven systems. In Proceedings of the 1st ACM Workshop on Privacy in the Electronic Society, pages 85–88.

Fatoretto, M. B., Özbulak, G., Anjos, A., and Berton, L. (2025). Optimizing fairness and utility in healthcare machine learning models. Applied Soft Computing, 181:113426.

Hardt, M., Price, E., and Srebro, N. (2016). Equality of opportunity in supervised learning. Advances in Neural Information Processing Systems, 29.

Little, R. J. and Rubin, D. B. (2019). Statistical analysis with missing data, volume 333. John Wiley & Sons.

Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., and Galstyan, A. (2021). A survey on bias and fairness in machine learning. ACM Computing Surveys (CSUR), 54(6):1–35.

O’Neil, C. (2016). Weapons of math destruction: How big data increases inequality and threatens democracy. Crown.

Pessach, D. and Shmueli, E. (2022). Algorithmic fairness. IEEE Transactions on Knowledge and Data Engineering, 34(10):4824–4842.

Rabonato, R. T. and Berton, L. (2025). A systematic review of fairness in machine learning. AI and Ethics, 5(3):1943–1954.

Verma, S. and Rubin, J. (2018). Fairness definitions explained. In Proceedings of the international workshop on software fairness, pages 1–7.

Wecchi, M. and Berton, L. (2026). Algorithmic fairness and bias mitigation in financial artificial intelligence: scoping review. AI and Ethics, 6(4):409.