International Journal of Innovative Research in Computer and Communication Engineering
ISSN Approved Journal | Impact factor: 8.771 | ESTD: 2013 | Follows UGC CARE Journal Norms and Guidelines
| Monthly, Peer-Reviewed, Refereed, Scholarly, Multidisciplinary and Open Access Journal | High Impact Factor 8.771 (Calculated by Google Scholar and Semantic Scholar | AI-Powered Research Tool | Indexing in all Major Database & Metadata, Citation Generator | Digital Object Identifier (DOI) |
| TITLE | Dataset-Dependent Trade-offs in Bias Mitigation for General-Purpose Data Science Pipelines |
|---|---|
| ABSTRACT | Machine learning models used in lending, hiring, and criminal justice can produce systematically different outcomes for different demographic groups, even when the protected attribute itself is excluded from the training features. A number of bias mitigation techniques have been proposed at different stages of the modeling pipeline — before training, during training, or after a model has already produced predictions — but there is little agreement on which stage works best, and even less evidence on whether a method that performs well on one dataset carries over to another. This paper compares three approaches on two widely used fairness benchmarks: the UCI Adult Income dataset (protected attribute: sex) and the COMPAS recidivism dataset (protected attribute: race). The methods evaluated are an unmitigated logistic regression baseline, Reweighing as a pre-processing method, and Threshold Optimizer as a post-processing method, scored on accuracy, demographic parity difference, and equalized odds difference. On Adult Income, Threshold Optimizer reduced demographic parity difference from 0.1720 to 0.0147 but raised equalized odds difference from 0.0905 to 0.3498. On COMPAS, the same method reduced both demographic parity difference (0.1613 to 0.0368) and equalized odds difference (0.1208 to 0.0204) together. Reweighing nearly halved demographic parity disparity on Adult Income at a cost of 0.4 accuracy points, but had no measurable effect on COMPAS. These results indicate that a mitigation method's fairness profile is not a fixed property of the algorithm; it interacts with dataset-specific factors such as the strength of the correlation between the protected attribute and the outcome. The paper argues for dataset-specific evaluation before deployment and for reporting more than one fairness metric, since optimizing for one can quietly worsen another. |
| AUTHOR | KARANVIR KAUR, SAHANA M M Assistant Professor, Department of Master of Computer Applications, Koshys Institute of Management Studies, Bengaluru, India Assistant Professor, Department of Master of Computer Applications, Koshys Institute of Management Studies, Bengaluru, India |
| VOLUME | 188 |
| DOI | DOI: 10.15680/IJIRCCE.2026.1409015 |
| pdf/15_Dataset-Dependent Trade-offs in Bias Mitigation for General-Purpose Data Science Pipelines.pdf | |
| KEYWORDS | |
| References |
[1] S. Caton and C. Haas, "Fairness in Machine Learning: A Survey," ACM Computing Surveys, vol. 56, no. 7, pp. 1-38, 2024. [2] D. Pessach and E. Shmueli, "A Review on Fairness in Machine Learning," ACM Computing Surveys, vol. 55, no. 3, pp. 1-44, 2022. [3] M. Hort, Z. Chen, J. M. Zhang, M. Harman, and F. Sarro, "Bias Mitigation for Machine Learning Classifiers: A Comprehensive Survey," ACM Journal on Responsible Computing, vol. 1, no. 2, pp. 1-52, 2023. [4] K. Orphanou et al., "Mitigating Bias in Algorithmic Systems: A Fish-Eye View," ACM Computing Surveys, vol. 55, no. 5, pp. 1-37, 2022. [5] H. Liu, Y. Wang, W. Fan, X. Liu, Y. Li, S. Jain, Y. Liu, A. Jain, and J. Tang, "Trustworthy AI: A Computational Perspective," ACM Transactions on Intelligent Systems and Technology, vol. 14, no. 1, pp. 1-59, 2022. [6] R. Kohavi, "Scaling Up the Accuracy of Naive-Bayes Classifiers: A Decision-Tree Hybrid," in Proc. 2nd Int. Conf. Knowledge Discovery and Data Mining (KDD), 1996, pp. 202-207. [7] J. Angwin, J. Larson, S. Mattu, and L. Kirchner, "Machine Bias," ProPublica, May 2016. [8] J. Larson, S. Mattu, L. Kirchner, and J. Angwin, "How We Analyzed the COMPAS Recidivism Algorithm," ProPublica, May 2016. [9] F. Kamiran and T. Calders, "Data Preprocessing Techniques for Classification Without Discrimination," Knowledge and Information Systems, vol. 33, no. 1, pp. 1-33, 2012. [10] M. Hardt, E. Price, and N. Srebro, "Equality of Opportunity in Supervised Learning," in Proc. 30th Conf. Neural Information Processing Systems (NeurIPS), 2016, pp. 3315-3323. [11] R. K. E. Bellamy et al., "AI Fairness 360: An Extensible Toolkit for Detecting, Understanding, and Mitigating Unwanted Algorithmic Bias," IBM Journal of Research and Development, vol. 63, no. 4/5, pp. 4:1-4:15, 2019. [12] S. Bird et al., "Fairlearn: A Toolkit for Assessing and Improving Fairness in AI," Microsoft Research Technical Report MSR-TR-2020-32, 2020. |
