Predicting slope stability potential failure surface using machine learning algorithms

This study investigated the performance of machine learning models in predicting the FS and slip surface. The models considered include support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost) algorithms. The slope stability analysis data for training of machine learning algorithms were obtained through the limit equilibrium method. This includes various scenarios of dry and homogeneous slope cases, encompassing a range of slope geometries (height (H), slope ratio (v/h)), and soil shear strength parameters (soil unit weight (γ), cohesion (c), friction angle (ϕ)). According to the evaluation using Taylor’s chart metrics, including standard deviation, correlation determination (R²), and root-mean-square error (RMSE), the XGBoost algorithm demonstrated the best performance. Additionally, employing the SHapley Additive exPlanations (SHAP) methodology revealed the significance order of variables as v/h > H > c > ϕ > γ for the factor of safety (FS) and H > v/h > c > ϕ > γ for the slip surface.