Applications of Stacking/Blending ensemble learning approaches for evaluating flash flood susceptibility

Flash floods are a type of catastrophic disasters which cause significant losses life and property worldwide. In recent years, machine learning techniques have become powerful tools for evaluating flash flood susceptibility. This research applies stacking blending ensemble approaches to assess the potential in Jiangxi, China. Four base models – linear regression, K-nearest neighbours, support vector machine, random forest adopted build two models. All evaluated by three metrics (accuracy, true positive rate, area under receiver operating characteristic curve) compared with Bayesian approach. The results suggest that approach is superior all other models, has then been selected evaluate vulnerability catchments Jiangxi. derived maps susceptibility over half province, terms either or number catchments, prone floods, particular north, northeast south. These empirical findings can help develop plans disaster prevention control, as well improving public knowledge hazards.