Fair Feature Importance Scores for Interpreting Decision Trees
Abstract: Across various sectors such as healthcare, criminal justice, national security, finance, and technology, large-scale machine learning (ML) systems are being deployed to make critical data-driven decisions. Many have asked if we can and should trust these ML systems to be making these decisions. Two critical components are prerequisites for trust in ML systems: interpretability, or the ability to understand why the ML system makes the decisions it does, and fairness, which ensures that ML systems do not exhibit bias against certain individuals or groups. While both interpretability and fairness have garnered substantial attention in the ML literature, methods directly interpreting models in terms of fairness remain limited. This paper considers a popular interpretation for a widely used class of ML models: feature importance scores for decision trees and tree-based models. We introduce a novel Fair Tree Feature Importance Score to assess each feature's impact on fairness or bias in decision trees. Analogous to the mean decrease in impurity for trees, our score quantifies the mean increase (or decrease) in group bias, and extends to interpret tree-based ensembles or surrogates of complex ML systems. Through simulations and real examples on benchmark fairness datasets, we show the validity of our Fair Tree Feature Importance Score, offering meaningful interpretations for both tree-based ensembles and tree-based surrogates of other ML systems.
Submission Length: Regular submission (no more than 12 pages of main content)
Assigned Action Editor: ~Aurélien_Bellet1
Submission Number: 2186
Loading