Investigation of the possibility of landslide hazard mapping using the Random Forest algorithm (Case study: Sardarabad Watershed, Lorestan Province)

With respect to the ability of data analysis techniques, their applications in various engineering and geosciences disciplines have been expanded. In this study, the random forest algorithm has been used for landslide susceptibility mapping in the Sardarabad Watershed, Lorestan Province. Random forest is another popular and very efficient algorithm, based on model aggregation ideas, for both regression and classification problems. The method combines the idea of bagger with random feature selection. For this purpose, layers of slope, aspect, elevation, curvature, distance from the fault, distance from the river, distance from the road, rainfall, lithology and land use were prepared as the factors influencing landslide. Then, their maps were digitized in ArcGIS10.2 map-software. Then, sensitive areas to landslides were evaluated using adaptive random forest algorithms. Meanwhile, random forest algorithms were written in R software and finally, ROC curves were used for evaluating the models. Based on the obtained results in the study area, the accuracy of the random forest algorithm is 98.8%. Overall, the random forest algorithm indicates that lithology and distance to roads are the main factors on landslide occurrence. Overall, the random forest algorithm indicates that lithology and distance to roads are the main factors on landslide occurrence.