Slope stability prediction based on AutoML for multiple failure mechanisms

Slope stability prediction is one of the most critical tasks in geotechnical and transportation engineering projects. Accurate prediction of slope stability is of great significance for the initial design of slope projects and disaster prevention. Currently, prediction methods combining data-driven approaches with AutoML (Automated Machine Learning) have achieved substantial research results in predicting the Factor of Safety (FOS) of slopes. In order to explore the generalization ability of AutoML models in predicting different failure mechanisms, this paper, for the first time, uses the TPOT (Tree-based Pipeline Optimization Tool) model to predict the FOS of slopes with commonly occurring failure mechanisms such as circular failure, translational failure, and buckling failure in the mountainous and canyon areas of Western Sichuan, China. The reliability of the TPOT model on datasets with different failure mechanisms and feature dimensions was verified using the Feature-engine missing value indicator. The experimental results show that the TPOT model achieves an R2 of 0.984, 0.984, and 0.991, with RMSE values of 0.039, 0.054, and 0.039 for the actual engineering case test sets of the three failure mechanisms, respectively. When facing data missing situations, the R2 values were 0.941, 0.940, and 0.988, with RMSE values of 0.047, 0.099, and 0.046, still demonstrating strong predictive ability, thus validating the generalization ability and robustness of the TPOT model. Additionally, a Multi-type Factor of Safety Prediction System (MSSPS) was developed in this paper, featuring a simple and user-friendly graphical user interface (GUI) to enable field personnel to make rapid preliminary assessments of slope stability.