FairPut: A Light Framework for Machine Learning Fairness with LightGBM

Abstract

This is a holistic framework to approach fair prediction outputs at the individual and group level. This framework includes quantitative monotonic measures, residual explanations, benchmark competition, adversarial attacks, disparate error analysis, model agnostic pre-and post-processing, reasoning codes, counterfactuals, contrastive explanations, and prototypical examples. A number novel techniques are proposed in this framework, each of which could benefit from future examination.

Keywords: Fairness, Explainability, Machine Learning, Robustness, Python, FairPut

JEL Classification: O, C, J