Domain-informed explainable boosting machines for trustworthy lateral spread predictions

TL;DR

This paper introduces a domain-informed framework to enhance the physical consistency of Explainable Boosting Machines for lateral spread prediction, ensuring more reliable natural hazard modeling while maintaining accuracy.

Contribution

It proposes a novel method to incorporate domain knowledge into EBMs, correcting non-physical behaviors in hazard prediction models.

Findings

Improved physical consistency in model explanations

Maintained predictive accuracy with only 4-5% tradeoff

Validated on the 2011 Christchurch earthquake dataset

Abstract

Explainable Boosting Machines (EBMs) provide transparent predictions through additive shape functions, enabling direct inspection of feature contributions. However, EBMs can learn non-physical relationships that reduce their reliability in natural hazard applications. This study presents a domain-informed framework to improve the physical consistency of EBMs for lateral spreading prediction. Our approach modifies learned shape functions based on domain knowledge. These modifications correct non-physical behavior while maintaining data-driven patterns. We apply the method to the 2011 Christchurch earthquake dataset and correct non-physical trends observed in the original EBM. The resulting model produces more physically consistent global and local explanations, with an acceptable tradeoff in accuracy (4--5\%).