SimFair: A Unified Framework for Fairness-Aware Multi-Label Classification

Recent years have witnessed increasing concerns towards unfair decisions made by machine learning algorithms. To improve fairness in model decisions, various notions been proposed and many fairness-aware methods are developed. However, most of existing definitions focus only on single-label classification. Fairness for multi-label classification, where each instance is associated with more than one labels, still yet to establish. fill this gap, we study classification paper. We start extending Demographic Parity (DP) Equalized Opportunity (EOp), two popular notions, scenarios. Through a systematic study, show that data, because unevenly distributed EOp usually fails construct reliable estimate labels few instances. then propose new framework named Similarity s-induced (sγ -SimFair). This utilizes data similar when estimating particular label group better stability, can unify DP EOp. Theoretical analysis experimental results real-world datasets together demonstrate the advantage sγ -SimFair over tasks.