|
Abstract
|
The main aim of this study was to evaluate and compare the results of two data-mining algorithms including support vector machine (SVM) and logistic model tree (LMT) for shallow landslide modelling in Kamyaran county where located in Kurdistan Province, Iran. A total of 60 landslide locations were identified using different sources and randomly divided into a ratio of 70/30 for landslide modeling and validation process. After that, 21 conditioning factors, with a raster resolution of 20 m, based on the information gain ratio (IGR) technique were selected. Performance of the models was evaluated using area under the receiver-operating characteristic curve (AUROC), and also several statistical-based indexes. Results depicted that only eight factors including distance to river, river density, stream power index (SPI), rainfall, valley depth, topographic wetness index (TWI), solar radiation, and plan curvature were known more effective for landslide modeling using training data set. The results also revealed that the SVM model (AUROC = 0.882) outperformed and outclassed the LMT model (AUROC = 0.737). Therefore, analysis and comparison of the results showed that the SVM model by RBF function performed well for landslide spatial prediction in the study area. Eventually, the findings of this study can be useful for land-use planning, reducing the risk of landslide, and decision-making in areas prone to landslide.
|