Flood probability zonation using a comparative study of two well-known random forest and support vector machine models in northern Iran

The current study is aimed to zoning flood probability map in the Saliantapeh catchment is located in the Golestan Province. To this aim, two well-known data mining models namely Random Forest (RF) and Support Vector Machine (SVM) were applied due to their robust computational algorithm. Flood inventories were gathered through several field surveys using, local information and available organizational resources and corresponding map was created in the geographic information system. Reviewing several worldwide studies, 13 predisposing variables including proximity to stream, soil texture, lithological units, land use/cover, slope percent, elevation/DEM, slope aspect, plan curvature, profile curvature, stream power index and topographic wetness index were chosen and the corresponding maps were generated in the geographic information system. In this study, after preparing the predictor maps, SPSS software was used to analyze this data and testing Multi-collinearity. In order to evaluate models’ results the area under the receiver operating were used. Three different sample data sets (s1, s2, s3) including 70% for training and 30% for validation were randomly gathered to evaluate the robustness of the applied models. Results showed that the RF model with the area under curve value of 0.96 and robustness of 0,001 in validation step had better performance on flood probability zonation over the study area.