Framework of domain-informed explainable boosting machines for lateral spreading prediction

This project provides an implementation of an Explainable Boosting Machine using data collected to study earthquake-induced lateral spreading during the 2011 Christchurch earthquake in New Zealand. The dataset, originally compiled by Durante and Rathje (2022) and available on DesignSafe (PRJ-2998: Machine Learning Models for the Evaluation of the Lateral Spreading Hazard in the Avon River Area Following the 2011 Christchurch Earthquake), contains peak ground acceleration, groundwater depth, elevation, slope, and distance to rivers.

The project focus on predicting and interpreting lateral spreading hazard using explainable boosting machine that can be evaluated against established geohazard knowledge. The project includes curated data files and a reproducible notebook that demonstrates model training, interpretation of feature effects, identification of physically inconsistent behavior, and incorporation of domain knowledge into the trained model.

The Explainable Boosting Machine is an inherently interpretable learning approach that represents predictions as additive lookup tables rather than opaque model parameters. This structure enables engineers and researchers to directly compute predictions, verify feature contributions, and modify the model in a controlled and transparent manner without full retraining.

This dataset and notebooks are intended to benefit geotechnical engineers, earthquake hazard researchers, and machine-learning practitioners by supporting reproducible analysis, validation of interpretable predictive models, development of new hazard-assessment methods, and educational use in transparent artificial intelligence for natural hazards.