Feature Importance via Sets of Locally Performant Linear Models

Fatemeh Tohidian, Davin Hill, Aria Masoomi, Peter J. Castaldi, Jennifer Dy
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:1531-1539, 2026.

Abstract

Understanding the contribution of individual features to a model’s prediction is critical in applications such as medicine. While feature importance methods aim to quantify how much a feature contributes to a model’s accuracy, they often overlook heterogeneous patterns in the data and suffer from limited robustness. We propose $\ell\text{-MCR}$, a local feature importance method that identifies meaningful neighborhoods around a point of interest, regions where the model or data behavior is locally stable and interpretable. Within these neighborhoods, we estimate feature importance using Model Class Reliance (MCR), which offers robustness by considering the full set of near-optimal models. We also provide a consistency proof for reliably detecting such neighborhoods. Experiments on both synthetic and real-world datasets demonstrate that $\ell\text{-MCR}$ captures localized feature importance patterns that global approaches fail to detect.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-tohidian26a, title = { Feature Importance via Sets of Locally Performant Linear Models }, author = {Tohidian, Fatemeh and Hill, Davin and Masoomi, Aria and Castaldi, Peter J. and Dy, Jennifer}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {1531--1539}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/tohidian26a/tohidian26a.pdf}, url = {https://proceedings.mlr.press/v300/tohidian26a.html}, abstract = { Understanding the contribution of individual features to a model’s prediction is critical in applications such as medicine. While feature importance methods aim to quantify how much a feature contributes to a model’s accuracy, they often overlook heterogeneous patterns in the data and suffer from limited robustness. We propose $\ell\text{-MCR}$, a local feature importance method that identifies meaningful neighborhoods around a point of interest, regions where the model or data behavior is locally stable and interpretable. Within these neighborhoods, we estimate feature importance using Model Class Reliance (MCR), which offers robustness by considering the full set of near-optimal models. We also provide a consistency proof for reliably detecting such neighborhoods. Experiments on both synthetic and real-world datasets demonstrate that $\ell\text{-MCR}$ captures localized feature importance patterns that global approaches fail to detect. } }
Endnote
%0 Conference Paper %T Feature Importance via Sets of Locally Performant Linear Models %A Fatemeh Tohidian %A Davin Hill %A Aria Masoomi %A Peter J. Castaldi %A Jennifer Dy %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-tohidian26a %I PMLR %P 1531--1539 %U https://proceedings.mlr.press/v300/tohidian26a.html %V 300 %X Understanding the contribution of individual features to a model’s prediction is critical in applications such as medicine. While feature importance methods aim to quantify how much a feature contributes to a model’s accuracy, they often overlook heterogeneous patterns in the data and suffer from limited robustness. We propose $\ell\text{-MCR}$, a local feature importance method that identifies meaningful neighborhoods around a point of interest, regions where the model or data behavior is locally stable and interpretable. Within these neighborhoods, we estimate feature importance using Model Class Reliance (MCR), which offers robustness by considering the full set of near-optimal models. We also provide a consistency proof for reliably detecting such neighborhoods. Experiments on both synthetic and real-world datasets demonstrate that $\ell\text{-MCR}$ captures localized feature importance patterns that global approaches fail to detect.
APA
Tohidian, F., Hill, D., Masoomi, A., Castaldi, P.J. & Dy, J.. (2026). Feature Importance via Sets of Locally Performant Linear Models . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:1531-1539 Available from https://proceedings.mlr.press/v300/tohidian26a.html.

Related Material