Evaluation of flood susceptibility prediction based on a resampling method using machine learning

Abstract The largest recorded flood loss occurred in the study area 2013. This aims to examine resampling methods (i.e. cross-validation (CV), bootstrap, and random subsampling) improve performance of seven basic machine learning algorithms: Generalized Linear Model, Support Vector Machine, Random Forest (RF), Boosted Regression Tree, Multivariate Adaptive Splines, Mixture Discriminate Analysis, Flexible Discriminant found factors causing flooding strongest correlation between variables. model is evaluated using Area Under Curve, Correlation, True Skill Statistics, Deviance. methodology was applied Kendari City, an urban that faced destructive floods. evaluation results show CV-RF has a good predicting susceptibility this with values, AUC = 0.99, COR 0.97, TSS 0.90, deviance 0.05. A total 89.44 km2 or equivalent 32.54% flood-prone dominant lowland morphology. Among 17 parameters cause flooding, strongly influenced by vegetation density index Terrain Roughness Index (TRI) 28 models. occurs TRI Sediment Transport (STI) 0.77, which means elements violence.