Application of support vector machines in landslide susceptibility assessment for the Hoa Binh province (Vietnam) with kernel functions analysis

The main objective of this study is to investigate the potential application of support vector machines (SVM) with kernel functions analysis for spatial prediction of landslides in the Hoa Binh province, Vietnam. A landslide inventory map that accounts for landslides that occurred during the last ten years was constructed using data from various sources. The landslide inventory was randomly divided into a training dataset 70% for building the models and the remaining 30% for the validation of the models. Ten landslide conditioning factors, such as slope angle, aspect, relief amplitude, lithology, soil type, landuse, distance to roads, distance to rivers, distance to faults and rainfall were prepared. During the model building process, four different SVM kernel functions (linear, polynomial, radial basic function, and sigmoid) were employed and four landslide susceptibility maps were constructed. Using the prediction rate method, the validation was performed by using landslide locations, which were not utilized during the model building. The validation results showed that the area under the curve (AUC) for landslide susceptibility maps produced by the SVM linear function, SVM polynomial function, SVM radial basic function, and SVM sigmoid function are 0.956, 0.956, 0.952, and 0.945 respectively. It indicates that the four landslide models seem to have performed well. Compared with the logistic regression (AUC =0.938) and Bayesian neural network model (AUC 0.903), the accuracy of the SVM landslide models in this study (using radial basic function and polynomial function) are slightly better. The result shows that SVM is a powerful tool for landslide susceptibility mapping at a regional scale. These maps can be very useful for natural hazards assessment and for land use planning.