SAILS: Surrogate-based Analysis of Interactions via Local Effect Smooths
Abstract
Feature interactions drive much of the predictive power of machine learning models, yet existing explanation methods only detect and quantify interactions without revealing their functional form, or visualize only restricted interaction types. We propose Surrogate-based Analysis of Interactions via Local Effect Smooths (SAILS), a model-agnostic framework that analyzes pairwise interactions through generalized additive model (GAM) surrogates fitted to the local effects of a black-box model. For each interval of a feature of interest, the surrogate smooth terms isolate the interaction components on derivative level, enabling (i) interaction detection through a heuristic derived from significance tests on smooth terms, (ii) interaction form categorization into linear, product-separable, and non-product-separable types, and (iii) tailored, interpretable visualizations for each interaction type. We empirically validate the framework through controlled simulations and a real-world task, showing its effectiveness for pairwise interactions, with limitations under strong feature correlations and higher-order interactions. SAILS fills a notable gap in the eXplainable AI (XAI) toolbox, going beyond detecting interactions alone to characterizing their functional form.
Related
- A Fractional Logistic-Type Elliptic Problem with Global Interactions: Existence, Local Uniqueness, and the Fractional-to-Local Limit
- Local Density Approximation and Other Limit Regimes for a Homogeneous Bose Gas with Repulsive Three-Body Interactions in Low-Dimensional Space
- Local determination of low-energy states in gapped fermion systems with weak long-range interactions
- Predictions as Surrogates: Revisiting Surrogate Outcomes in the Age of AI
- Recommendations for Evaluating A Single Surrogate Endpoint (REASSurE): Consensus-based surrogate evaluation framework with explanation and elaboration
BibTeX
@misc{heiß2026sails,
title = {{SAILS: Surrogate-based Analysis of Interactions via Local Effect Smooths}},
author = {Timo Heiß and Julia Herbinger and Bernd Bischl and Giuseppe Casalicchio},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2606.09404},
eprint = {2606.09404},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.09404},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - SAILS: Surrogate-based Analysis of Interactions via Local Effect Smooths
AU - Timo Heiß
AU - Julia Herbinger
AU - Bernd Bischl
AU - Giuseppe Casalicchio
PY - 2026
DA - 2026/09/13
PB - arXiv
DO - 10.48550/arxiv.2606.09404
UR - https://arxiv.org/abs/2606.09404
AB - Feature interactions drive much of the predictive power of machine learning models, yet existing explanation methods only detect and quantify interactions without revealing their functional form, or visualize only restricted interaction types. We propose Surrogate-based Analysis of Interactions via Local Effect Smooths (SAILS), a model-agnostic framework that analyzes pairwise interactions through generalized additive model (GAM) surrogates fitted to the local effects of a black-box model. For each interval of a feature of interest, the surrogate smooth terms isolate the interaction components on derivative level, enabling (i) interaction detection through a heuristic derived from significance tests on smooth terms, (ii) interaction form categorization into linear, product-separable, and non-product-separable types, and (iii) tailored, interpretable visualizations for each interaction type. We empirically validate the framework through controlled simulations and a real-world task, showing its effectiveness for pairwise interactions, with limitations under strong feature correlations and higher-order interactions. SAILS fills a notable gap in the eXplainable AI (XAI) toolbox, going beyond detecting interactions alone to characterizing their functional form.
ER -