Chi-square automatic interaction detection (CHAID) algorithm for flood susceptibility assessment in Sardabroud watershed, Iran

Abstract

Flood, as a natural phenomenon, is the most common natural hazard that causes significant damage in the world. It is difficult to predict and identify flood zones due to variable weather conditions and various influencing factors. However, the identification and detection of early flood zones using machine learning techniques is used for smart flood management. In this study, Chi-square automatic interaction detection (CHAID) machine learning model for flood susceptibility map in Sardabroud watershed in north of Iran has been evaluated. For this purpose, a spatial database including 205 present and past flood locations with 8 conditional factors including elevation, slope, landuse, normalized difference vegetation index (NDVI), distance to river, topographic wetness index (TWI), lithology and rainfall are considered. After calculating variance inflation factors (VIF), all of the flood factors were considered for the modeling process. VIF technique uses to quantify multi-collinearity. Receiver operating characteristic (ROC), area under curve (AUC) and accuracy (ACC) metrics were used to evaluate and compare the predictability of the model. The results show that the CHAID model reaches an AUC of 0.939. This model has been proven as an efficient model for detecting flood prone areas in this watershed.