How Reliable are Fairness Audits with Unreliable Data?
A new paper explores the reliability of fairness audits in machine learning when data on protected attributes is incomplete. Researchers found that missing protected-label data often does not significantly alter the recommendations of common mitigation methods. However, threshold optimization can inadvertently lead to intersectional harm, even when fairness gains are observed on single axes. AI
IMPACT Highlights potential pitfalls in evaluating ML model fairness, urging caution in interpreting audit results with incomplete data.