Research on the predictability of rock strength under freeze-thaw cycles - A hybrid model of SHAP-IPOA-XGBoost

Abstract

As environmental dynamics increasingly influence the stability of engineering materials, accurately predicting sandstone strength under freeze-thaw cycles has become essential. This study presents a new method, IPOA-XGBoost, integrating an improved Pelican Optimization Algorithm (IPOA) with Extreme Gradient Boosting (XGBoost) to accurately forecast sandstone strength in freeze-thaw environments. Principal component analysis (PCA) is employed for reducing data dimensionality, while IPOA optimizes the hyperparameters of the XGBoost model. Experimental findings show that the IPOA-XGBoost model outperforms traditional methods, delivering improved accuracy and robustness in both training and testing datasets. To address the “black box” challenge of machine learning models, SHAP (SHapley Additive exPlanations) values are applied to clarify the impact of individual features on prediction outcomes, validating SHAP's reliability as an interpretive tool. The findings highlight the importance of strain rate (SR), impact pressure (IP), and confining pressure (CP) as key variables affecting sandstone strength predictions. This methodology provides significant insights for predicting sandstone strength in applied engineering contexts.