Whence Is A Model Fair? Fixing Fairness Bugs via Propensity Score Matching
This work addresses fairness evaluation and mitigation for machine learning models, particularly in social applications, but is incremental as it builds on existing post-processing and propensity score matching techniques.
The paper tackles the problem of unreliable fairness metrics in machine learning due to biased sampling between training and test data, proposing FairMatch, a post-processing method that uses propensity score matching to identify and mitigate bias, which significantly reduces bias on affected data subsets without sacrificing predictive performance.
Fairness-aware learning aims to mitigate discrimination against specific protected social groups (e.g., those categorized by gender, ethnicity, age) while minimizing predictive performance loss. Despite efforts to improve fairness in machine learning, prior studies have shown that many models remain unfair when measured against various fairness metrics. In this paper, we examine whether the way training and testing data are sampled affects the reliability of reported fairness metrics. Since training and test sets are often randomly sampled from the same population, bias present in the training data may still exist in the test data, potentially skewing fairness assessments. To address this, we propose FairMatch, a post-processing method that applies propensity score matching to evaluate and mitigate bias. FairMatch identifies control and treatment pairs with similar propensity scores in the test set and adjusts decision thresholds for different subgroups accordingly. For samples that cannot be matched, we perform probabilistic calibration using fairness-aware loss functions. Experimental results demonstrate that our approach can (a) precisely locate subsets of the test data where the model is unbiased, and (b) significantly reduce bias on the remaining data. Overall, propensity score matching offers a principled way to improve both fairness evaluation and mitigation, without sacrificing predictive performance.