Rapid prediction of landslide dam stability considering the missing data using XGBoost algorithm

The stability prediction of landslide dams is the basis for reducing and even eliminating the damage caused by landslide dam failures. However, the stability of landslide dams with missing data cannot be predicted by the existing rapid evaluation methods. A landslide dam database containing 2783 historical cases from 49 countries worldwide is created. Considering 8 variables and the missing data, the XGBoost model for predicting the stability of landslide dams is established, and the impact of missing data on the prediction accuracy of the model is also discussed. The XGBoost prediction model can not only predict the landslide dam cases with missing data, but the accuracy of the XGBoost model has also been improved compared with that of the widely used rapid evaluation methods. The characteristic analysis framework of landslide dam stability prediction is proposed based on SHapley Additive exPlanations attribution theory, and the results show the importance of each characteristic and its influence on landslide dam stability from both overall and individual perspectives. The Meilonggou, Usoi, and Mount St. Helens landslide dams are introduced to demonstrate the applicability and accuracy of the XGBoost prediction model proposed in this study.