This study explores the prediction of macroseismic intensity (MI) from ground motion parameters through a comparative evaluation of multiple regression models. A dataset comprising 296 sets of seismic records, each annotated with investigated MI values and 42 ground motion parameters, was constructed. A wide range of regression approaches—including linear models, tree-based methods, and ensemble learning algorithms were benchmarked to identify the optimal model in terms of predictive accuracy and robustness. Among these, XGBoost consistently outperformed other methods across various feature subsets, demonstrating superior accuracy and resilience to dimensionality reduction. To enhance model interpretability, SHAP (SHapley Additive exPla-nations) analysis was employed, identifying Formula: see text (vertical peak ground velocity), Formula: see text (the average value of acceleration response spectrum at 2.6 s), DI (destructiveness index), and Formula: see text (significant duration between 5% and 95% of Arias intensity) as the most influential predictors. These findings underscore the dual advantages of XGBoost in delivering both high predictive performance and interpretable insights, offering practical implications for seismic hazard assessment and intensity based early warning systems.
Zhang et al. (Fri,) studied this question.