Development of machine learning surrogate models for slope stability prediction using AI techniques: a case study of...

Rainfall-induced landslides present a critical global geohazard, necessitating the development of robust, rapid tools for slope stability evaluation. This study proposes a hybrid framework that integrates numerical modeling with machine learning (ML) regression to predict the Factor of Safety (FoS) under dynamic groundwater conditions. Using the geometry of the recent Meizhou landslide in China as a baseline, a parametric study was conducted via GeoStudio's limit equilibrium analyses to generate a dataset of 249 simulations based on five key geotechnical parameters: cohesion, friction angle, unit weight, surcharge load, and groundwater level. Three regression-based ML models such as Random Forest (RF), Ordinary Least Squares (OLS), and Extreme Gradient Boosting (XGBoost) were trained to develop interpretable surrogate equations. A novel post-regression linear calibration method was applied to minimize residual errors and enhance the alignment of predicted versus actual FoS values. The results demonstrate that XGBoost achieved the highest predictive accuracy , effectively capturing complex nonlinear relationships. Notably, the Random Forest model exhibited the most significant performance gain from the calibration process. This study establishes practical, high-precision surrogate equations suitable for AI-augmented geotechnical assessments, offering a reliable solution for real-time safety prediction in hydrologically active slopes.