ESENSC paper proposes linear-cost alternative to SHAP feature attribution
A new arXiv paper introduces ESENSC, a deterministic attribution rule that needs only linearly many coalition-value evaluations and reports close agreement with exact SHAP on tabular tasks.
In plain words
Machine learning models often explain predictions by scoring how much each input feature mattered, and SHAP is a popular but computationally expensive way to do this. The authors of ESENSC propose a cheaper formula that, in their tests, gives similar feature scores while doing far fewer calculations.
What was announced
The paper, posted on arXiv, presents ESENSC as a deterministic, closed-form attribution rule that satisfies the null-player property and is characterized axiomatically through efficiency, a restricted form of differential marginality, and explicit computational constraints. Where exact SHAP requires exponentially many coalition evaluations, the authors state that ESENSC scales linearly with the number of features.
Empirically, the authors report that on tabular prediction tasks using neural network and XGBoost models, ESENSC closely matched exact SHAP in attribution magnitudes and feature rankings while needing substantially less computation, and that under the evaluated settings it compared favorably with widely used sampling-based SHAP approximations. The evidence comes from the paper's abstract alone, so the experimental details, model configurations and any peer review status cannot be verified from the supplied material.
What it does not show
- The evidence is limited to the paper's abstract; experimental details, code availability and peer review status are not covered, and the reported agreement with SHAP has not been independently confirmed.
Sources
Source-support review completed 17 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 17 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.