February 1, 2021 Download

Using Shapley Values to Enable Machine Learning-driven Adverse Action

Written by Joseph Hammond, Aaron McGuire, Xuetong Li, Vachagan Darbinyan

This paper lays out best practices we have developed to better leverage Shapley values in enabling ML model deployment in credit issuance. Through a variety of analyses and comparisons to existing Adverse Action methodology, we demonstrate how Shapley values can be used to satisfy the Adverse Action requirements and break down a common barrier that lenders face in adopting more sophisticated ML models.