Algorithmic Fairness
本文综述算法公平性文献，重点讨论信用评分中的偏差度量与公平性指标，并应用于美国抵押贷款数据，发现性别和少数族裔群体存在不平衡，机器学习优于逻辑回归但解释性差。
This article reviews the recent literature on algorithmic fairness, with a particular emphasis on credit scoring. We discuss human versus machine bias, bias measurement, group versus individual fairness, and a collection of fairness metrics. We then apply these metrics to the US mortgage market, analyzing Home Mortgage Disclosure Act data on mortgage applications between 2009 and 2015. We find evidence of group imbalance in the dataset for both gender and (especially) minority status, which can lead to poorer estimation/prediction for female/minority applicants. Loan applicants are handled mostly fairly across both groups and individuals, though we find that some local male (nonminority) neighbors of otherwise similar rejected female (minority) applicants were granted loans, something that warrants further study. Finally, modern machine learning techniques substantially outperform logistic regression (the industry standard), though at the cost of being substantially harder to explain to denied applicants, regulators, or the courts.