Finding representative group fairness metrics using correlation estimations

Hadis Anahideh, Nazanin Nezami, Abolfazl Asudeh

Abstract: Highlights•Proposed a framework to estimate correlation among fairness notions.•Monte Carlo sampling used for efficient perturbation in model space.•Identifies a set of diverse metrics as representative of a given context.•Alleviates burden of choice of fairness, generalizes to arbitrary set of metrics.•Validated on real-world benchmark datasets.