An Empirical Study on Learning Fairness Metrics for COMPAS Data with Human Supervision

TL;DR

This paper empirically investigates how human supervision can be used to learn similarity metrics for individual fairness in criminal recidivism prediction, demonstrating that learned metrics outperform standard metrics.

Contribution

It introduces a new dataset of human judgments on COMPAS data and evaluates metric learning methods to derive similarity metrics aligned with human fairness perceptions.

Findings

Learned metrics outperform Euclidean and Precision metrics

Human supervision can effectively guide similarity metric learning

Empirical results support using human judgments for fairness metric development

Abstract

The notion of individual fairness requires that similar people receive similar treatment. However, this is hard to achieve in practice since it is difficult to specify the appropriate similarity metric. In this work, we attempt to learn such similarity metric from human annotated data. We gather a new dataset of human judgments on a criminal recidivism prediction (COMPAS) task. By assuming the human supervision obeys the principle of individual fairness, we leverage prior work on metric learning, evaluate the performance of several metric learning methods on our dataset, and show that the learned metrics outperform the Euclidean and Precision metric under various criteria. We do not provide a way to directly learn a similarity metric satisfying the individual fairness, but to provide an empirical study on how to derive the similarity metric from human supervisors, then future work can use this as a tool to understand human supervision.