# Understanding Global Feature Contributions With Additive Importance Measures

arXiv: https://arxiv.org/abs/2004.00668 · HTML: https://mlsanity.com/paper/2004.00668/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 1 Apr 2020.

## Built on by (most cited first, 10 of 10)

- [Do Feature Attribution Methods Correctly Attribute Features?](https://mlsanity.com/paper/2104.14403/) (2021)
- [Towards Trustworthy and Aligned Machine Learning: A Data-centric Survey with Causality Perspectives](https://mlsanity.com/paper/2307.16851/) (2023)
- [Feature Importance: A Closer Look at Shapley Values and LOCO](https://mlsanity.com/paper/2303.05981/) (2023)
- [SHAFF: Fast and consistent SHApley eFfect estimates via random Forests](https://mlsanity.com/paper/2105.11724/) (2021)
- [MDA for random forests: inconsistency, and a practical solution via the Sobol-MDA](https://mlsanity.com/paper/2102.13347/) (2021)
- [Accurate Shapley Values for explaining tree-based models](https://mlsanity.com/paper/2106.03820/) (2021)
- [Stabilizing Estimates of Shapley Values with Control Variates](https://mlsanity.com/paper/2310.07672/) (2023)
- [Confident Feature Ranking](https://mlsanity.com/paper/2307.15361/) (2023)
- [Greedy Modality Selection via Approximate Submodular Maximization](https://mlsanity.com/paper/2210.12562/) (2022)
- [Regional Explanations: Bridging Local and Global Variable Importance](https://mlsanity.com/paper/2604.11223/) (2026)

## Abstract (arXiv)

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.