Evaluation of the indirect tensile strength (ITS) of cement-treated clayey soils using XGBoost prediction model

Abstract

This study develops a data-driven framework using Extreme Gradient Boosting (XGBoost) to predict the Indirect Tensile Strength (ITS) of cement-treated clayey soils. Using 180 specimens with five input variables - cement content, curing time, curing temperature, plasticity index, and compaction energy - the model was trained (80%) and tested (20%), achieving strong accuracy (training R²=0.932, RMSE=55.41 kPa; testing R²=0.922, RMSE=81.98 kPa). SHapley Additive exPlanations (SHAP) analysis was applied for interpretability, showing cement content (39.3%) and curing time (30.1%) as the dominant predictors, followed by plasticity index (12.9%), while compaction energy (9.1%) and curing temperature (8.6%) had minor influence. K-means clustering combined with SHAP waterfall plots identified five distinct strength-development behavior groups, offering mechanistic insight into ITS variability. Overall, the XGBoost-SHAP framework proves to be a robust, interpretable tool for performance-based design and mixture optimization of cement-treated soils in infrastructure applications.