Domain-informed explainable boosting machines for trustworthy lateral spread predictions
This work addresses reliability issues in natural hazard applications, such as earthquake predictions, by enhancing the physical consistency of machine learning models, though it is incremental as it builds on existing EBM methods.
The study tackled the problem of non-physical relationships in Explainable Boosting Machines (EBMs) for lateral spreading predictions by introducing a domain-informed framework to modify shape functions, resulting in more physically consistent explanations with a 4-5% accuracy tradeoff.
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\%).