Modeling of Landslide Susceptibility Mapping Using State-Of-Art Machine Learning Models

Abstract

This paper presents the modeling of landslide susceptibility mapping of Penang Island, Malaysia using the state-of-art machine learning models. Machine learning models employed to generate landslide susceptibility maps are artificial neural network (ANN), extreme gradient boosting (XGBOOST), support vector machine (SVM), and logistic regression (LR). The effects and contributions of the landslide influencing factors that cause landslides are determined using Pearson’s and distance correlation coefficients. Prior to the training phase of the models, these landslide factors are processed using normalization and principal component analysis (PCA) to improve the prediction ability. The comprehensive performance of the models are evaluated with classification accuracy and receiver operating characteristics (ROC) curve. The obtained results of ROC indicate that ANN model, which has an accuracy of 96.43%, is the most accurate method for predicting the occurrence of landslides in Penang Island. It is followed by SVM (91.05%), XGBOOST (90.86%), and LR (80.05%).