Understanding Global Feature Contributions With Additive Importance Measures
This addresses the need for better global feature importance measures in machine learning, though it is incremental as it builds on existing literature.
The paper tackled the problem of global interpretability in complex machine learning models by introducing SAGE, a model-agnostic method that quantifies feature importance while accounting for interactions, and experiments showed it assigns more accurate importance values than other methods.
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.