Understanding Global Feature Contributions With Additive Importance Measures

TL;DR

This paper introduces a unified framework for global feature importance in machine learning models, proposing a new method called SAGE that accurately quantifies feature contributions considering interactions.

Contribution

It formalizes additive importance measures, unifies existing methods, and introduces SAGE, a model-agnostic approach that improves accuracy and efficiency in feature importance estimation.

Findings

SAGE can be computed efficiently.

SAGE provides more accurate importance values.

The framework unifies various importance methods.

Abstract

Understanding the inner workings of complex machine learning models is a long-standing problem and most recent research has focused on local interpretability. To assess the role of individual input features in a global sense, we explore the perspective of defining feature importance through the predictive power associated with each feature. We introduce two notions of predictive power (model-based and universal) and formalize this approach with a framework of additive importance measures, which unifies numerous methods in the literature. We then propose SAGE, a model-agnostic method that quantifies predictive power while accounting for feature interactions. Our experiments show that SAGE can be calculated efficiently and that it assigns more accurate importance values than other methods.