Estimating ML-driven compressive strength of geo-polymer stabilized soils

Predicting the Unconfined Compressive Strength (UCS) of stabilized soils is critical for sustainable ground improvement. Experimental investigations have been performed for various cases of geopolymer-stabilized soils, both in academic and professional contexts. However, a rational and data-driven prediction of UCS—a key parameter for design—remains a major concern. This study explores several Machine Learning (ML) approaches to estimate the UCS of geopolymer-stabilized soils using compiled experimental data. Eight influential input parameters were considered to represent soil properties, geopolymerization parameters, and mix design parameters. After evaluating feature significance, five ML models—Linear Regression (LR), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and Gradient Boosting (GB)—were trained and tested for UCS prediction. Among these, GB achieved superior performance (Test R² = 0.972, Test RMSE = 0.203 MPa), followed by SVM (Test R² = 0.898, Test RMSE = 0.387 MPa) and RF (Test R² = 0.898, Test RMSE = 0.387 MPa). LR (Test R² = 0.702, Test RMSE = 0.661 MPa) and DT (Test R² = 0.895, Test RMSE = 0.393 MPa) exhibited comparatively lower predictive capability. To enhance model interpretability, SHAP (SHAPley Additive exPlanations) analysis was employed, providing insights into the contribution and influence of individual input features on UCS predictions. The results establish GB as a highly reliable model for predicting the UCS of geopolymer-stabilized soils, while SHAP-based interpretability aids in understanding the key factors influencing strength development, thereby contributing to optimized mix design and sustainable geotechnical engineering applications.