Machine Learning–Based Prediction of Grouting Volume for Postgrouted Piles in the Lower Yellow River Region

Based on field construction data from postgrouted piles in the Jining–Liangshan Expressway Flood Detention Area and the Beijing–Hangzhou Grand Canal Bridge Project, this study presents a statistical analysis of grouting volumes at pile sides and tips across various soil layers in the lower Yellow River region. Empirical grouting coefficients tailored to this geological setting were proposed. Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) were employed to develop multifactor predictive models for grouting volume, with hyperparameters optimized using a K-fold Bayesian optimization framework to enhance model robustness and prediction performance. The influence of feature parameters and their interactions was quantitatively examined using SHapley Additive exPlanations (SHAP). Validation through two engineering case studies confirmed the applicability and accuracy of both the empirical coefficients and the machine learning models. Results demonstrate that the machine learning models significantly outperform empirical methods in predicting grouting volume, with XGBoost exhibiting the best overall performance. While SVR showed limitations due to data clustering and RF displayed larger deviations in data-scarce soil layers, SHAP interaction analysis revealed critical synergistic effects between key parameter combinations, such as silty clay with side grouting and groundwater depth with grouting depth. This study recommends XGBoost for predicting grouting volume in postgrouted piles in the lower Yellow River region, while the code-specified method incorporating the proposed empirical coefficients serves as a reliable supplementary approach for preliminary engineering estimations.