An interpretable evolutionary extreme gradient boosting algorithm for rock slope stability assessment

Globally, slope failures cause severe disasters and substantial financial losses annually. Recent advancements in machine learning (ML) algorithms and dataset collection have created alternate solutions for complex slope stability problems. However, rock slope stability prediction remains a challenging problem due to factors such as inadequate data and insufficient generalization performance of rock slope prediction models. The black-box nature of AI models also causes further criticism of using such models to address issues such as slope stability. In this study, we proposed an artificial intelligence (AI) based technique for rock slope stability prediction based on evolutionary and ML algorithms. The proposed GA-XGBoost model uses XGBoost to model the relationship between the input and output parameters of rock slopes, while Genetic Algorithm (GA) optimizes the hyperparameters of XGBoost. A comprehensive rock slope database of 7525 slope cases is implemented in this study to develop and verify the model. The model attains an impressive performance score of R 2 = 0.9999 , M A E = 0.8006 , and R M S E = 1.8624 on the training dataset and R 2 = 0.9934 , M A E = 2.2793 , and R M S E = 11.1090 on the testing dataset. Furthermore, to assess the relative significance of the various influential slope parameters, the SHapley Additive exPlanations (SHAP) algorithm is implemented. This step enables the physical and quantitative interpretations of dependencies between the input and output variables. Generally, this relationship is hidden in traditional machine learning algorithms.