FairShap: A Fairness Framework Based Explainable Machine Learning

With the increasing application of machine learning in real-world decision-making systems, the fairness and interpretability of tasks involving humans have not yet been fully guaranteed. In order to solve the above problems, we propose an interpretable fairness framework based on feature contributions, which aims to improve the degree of interpretability of fairness in binary classification tasks. First, the fairness contribution is explained by the importance of interpretable features and quantified by the Shapley value in Game Theory; then, groups are divided according to different protective attributes, and discrimination detection and debiasing algorithms are applied to specific groups to mitigate the bias in the original samples. The experimental results show that the proposed method significantly outperforms the existing methods in terms of interpretability and demonstrates wide applicability to different classifiers and fairness metrics.