Landslide-Prediction-2008-Wenchuan-Earthquake-Sichuan-China

Abstract.The objective of machine learning study for landslides are to predict the occurrence of landslide events as well as to identify several parameters which influence the landslide susceptibility. The study area is in the Northern End of the Longmenshan Fault, Sichuan Province, China. In this study, the authors used both the IBM SPSS program and the python language program to analyze and model the data provided. In data preparation, several steps are performed including data cleaning, aspect transformation, lithology separation, factor of safety (fos) classification, transformation, and normalization. To test the model accuracy, data partition was applied. The main modeling technique performed with SPSS software including Decision Tree, Support Vector Machine (SVM), Neural Network and KNN with additional logistic regression, K-Means, and PCA which run in python. Models are evaluated by using the confusion matrix with precision, recall, f1 score and accuracy to help determine the most suitable model. In general, all models with an accuracy score of more than 0.8, could be used to analyzed landslide susceptibility. Based on the analysis of data provided; scarps distance, specific weight, and lithology of moraine play the major roles for landslide susceptibility. Keywords: Machine Learning, Landslide Susceptibility, Confusion Matrix, Scarps Distance