A Study on Development of Landslide Susceptibility Map in Malaysia Landslide Prone Areas by Using Geographic Information System (GIS) and Machine Learning

Abstract

Landslide is a natural disaster that is common and frequently occurring in Malaysia. Thus, to reduce the impact of the landslide’s tragedy, a landslide susceptibility map is needed. The ultimate goal of this paper is to use Geographic Information System (GIS) and machine learning to develop a landslide susceptibility map. In two different landslide-prone areas in Malaysia, the performance of the two different machine learning models, Random Forest and Extreme Gradient Boosting (XGBoost) are evaluated and cross-validated. The Cameron Highland and Penang Island, Malaysia which are the subjects of this study, have a total of 233 and 443 landslides locations, respectively. These landslide locations were randomly divided into 70% for training and 30% for testing. The Digital Elevation Model (DEM), slope angle, slope length, Normalized Vegetation Index (NDVI), plan curvature, profile curvature, distance from the stream, distance from roads, Topographic Wetness Index (TWI) and Stream Power Index (SPI) are among the ten landslide conditioning factors, for which the spatial databases were developed by using GIS software. The area under the curve (AUC) of the receiver operating characteristic curve (ROC) had been applied to evaluate the machine learnings prediction accuracy. The result indicated that both XGBoost and Random Forest had a great performance across both study areas. For Penang Island, the AUC of XGBoost is 95.02% and the AUC of Random Forest is 94.99%. Meanwhile, for Cameron Highland, the AUC of XGBoost is 91.99% and the AUC of Random Forest is 92.32%. The final prediction map from this study might be useful for better planning in mitigating the occurrence of landslides.