ABSTRACT
Landslide Dam (LD) is a kind of natural hazards that occurs worldwide. Typically, LDs are structurally weak and prone to failure in a short time after formation, posing a threat to downstream areas. Therefore, it is crucial to have a method for quickly determining the stability of an LD. However, traditional research methods, such as model testing and numerical calculations, are time-consuming and may not meet this requirement. In this paper, we trained a Support Vector Machine (SVM) based prediction/classification model to analyze the stability of LDs. Since the success of data-driven AI approaches highly depends on the number and type of data, we discuss the selection of training datasets and input parameters by utilizing a real-world LD dataset. Experimental results demonstrated that the prediction accuracy of LD stability can be improved by incorporating hydraulic information along with geometric information, compared to using geometric information alone.
