Machine learning models for predicting soil shear strength using geotechnical parameters and explainable artificial intelligence

Accurate prediction of soil shear strength is essential for the design and stability assessment of geotechnical structures. In this study, machine learning (ML) techniques were employed to predict soil shear strength using seven geotechnical parameters: cohesion, angle of internal friction, liquid limit (LL), plastic limit (PL), plasticity index (PI), maximum dry density (MDD), and optimum moisture content (OMC). A dataset comprising 229 laboratory records of experimentally measured shear strength obtained from direct shear tests was divided into 80% training and 20% testing subsets. Four ML models, namely Linear Regression (LR), Decision Tree (DT), Random Forest (RF), and XGBoost (XGB), were developed and evaluated using multiple performance metrics and repeated cross-validation. On the independent testing dataset, RF provided the best overall performance, achieving an R 2 of 0.7812, adjusted R 2 of 0.7408, RMSE of 22.7080, and MAPE of 0.1523. XGB showed comparable performance, while DT exhibited the weakest generalization. SHAP, FAST, and PAWN analyses indicated that cohesion was the dominant model input for LR, DT, and RF, whereas OMC showed the highest sensitivity for XGB. The target shear strength values were experimentally obtained from direct shear tests rather than calculated from cohesion and angle of internal friction. However, the observed strong dependence between cohesion and shear strength, together with correlations among several input variables, limits the interpretation of the models as fully independent predictors. Overall, the results demonstrate the usefulness of interpretable ML models for representing laboratory-derived shear strength relationships under the investigated conditions, while further validation using independent datasets and wider stress conditions is required before broader application.