Bayesian ensemble learning and Shapley additive explanations for fast estimation of slope stability with a physics-informed database

Slope failures present substantial threats to public safety and economic losses. However, it remains challenging to achieve satisfactory performance due to insufficient datasets with machine learning (ML)-based slope stability assessment. In this study, an expanded physics-informed dataset was constructed by integrating historical case studies with data derived from nonintrusive stochastic analysis. The Bayesian ensemble learning model was employed to enhance prediction accuracy, with the Shapley additive explanations method employed to elucidate the contribution of each input variable. The proposed method displayed satisfactory performance, achieving an area under the curve of 0.9973, accuracy of 0.9727, and F1-score of 0.9729, surpassing the compared ML methods. Its robustness and generalization capabilities were confirmed through evaluations on diverse datasets and random seeds. Furthermore, a user-friendly graphical user interface was created for fast estimation of slope stability (FESS) using the trained prediction model. The performance of FESS was validated on a series of examples including the Australian Association for Computer-Aided Design referenced slope example EX1 and 77 in situ cases. This tool offers practitioners a high-performance solution, significantly reducing the effort required for slope stability assessments.