Assessment of flood-risk areas using random forest techniques: Busan Metropolitan City

Climate change increases both the risks and effects of flooding in urban areas, which, without mitigation, may lead to social catastrophes. In Korea, devastating typhoons and overflows account for approximately 90% of the country’s natural disasters, and the many man-made features of urban environments exacerbate the detrimental effects whenever flooding occurs. Many regression analysis methods exist for assessing geographical flood risk; furthermore, a handful of machine learning methods have been created for mitigation and estimation purposes—there are none for prevention. Therefore, in this study, we developed a machine learning flood assessment model that leverages several machine learning models for the Busan Metropolitan City. Each was applied to a test dataset, and their performances were evaluated based on accuracy, sensitivity, specificity, and area under the curve; thereafter, the model determined to be the most reliable was used to create a flood risk assessment map. The model was then used to assess the areas of highest probability of flooding. Upon its completion, we discovered that flooding may now occur with less rainfall than that of the 10-year return period. The derived map is expected to be used as a basic source for the development of preventive countermeasures against urban flooding, thus contributing to the enhancement of flood control and response capacities in applicable regions.