Landslide Susceptibility Mapping for Muzaffarabad Region Using Machine Learning Algorithms

In this study machine learning methods are proposed to identify landslides using data obtained from northern Pakistan (Muzaffarabad district). We used Random Forest, Decision Tree, gradient boosting machine (GBM), light gradient boosting machine (LGBM), neural network (NN), 1D-CNN (convolutional neural network one dimensional), XGBboost and Catboost for classification. Landslide is predicted with the dataset of 12 different factors and data of 1213 sites. First, we trained models individually and observed which one has the highest accuracy. Finally, the performances of applied models are assessed by using several statistical methods. Based on the results we obtained the highest accuracy is 85% of CNN and above 80% for boosting methods. The developed system can be effectively used to predict landslides of regions with similar environmental conditions.