Probabilistic prediction of rate‐dependent rock strength using natural gradient boosting and Gaussian process regression

The dynamic compressive strength () of rocks exhibits sensitivity to strain rate (), which is critical for understanding failure mechanisms in blasting, tunneling, and seismic response. Conventional empirical and deterministic models often struggle to generalize across diverse lithologies and loading conditions. This study develops an interpretable and probabilistic ensemble learning framework to predict using a curated dataset covering common rock types and experimental conditions incorporating static compressive strength (), , P‐wave velocity, bulk density, grain size, and specimen geometry. Ensemble models including random forest, extra tree, and extreme gradient boosting were benchmarked against uncertainty‐aware methods namely natural gradient boosting (NGBoost) and Gaussian process regression, with hyperparameters optimized via Bayesian optimization. NGBoost with decision tree base learner achieved the best test performance (Coefficient of determination = 0.954, root mean square error = 13.29 MPa, and mean absolute error = 8.59 MPa), demonstrating strong generalization. Model interpretability analyses (shapley additive explanations, permutation importance, and partial dependence) identified and as the most influential features, while geometric factors had minimal impact. The proposed framework provides a robust and interpretable tool for modeling rate‐dependent rock strength in dynamic geomechanical applications.