Flood forecasting using machine learning methods in a visual programming environment

Abstract

Floods are among the most destructive natural phenomena, significantly impacting the environment, economy, infrastructure, and public safety. Effective forecasting of such emergencies is crucial, especially in the context of global climate change and the increasing population density in coastal and low-lying areas. This work aims to develop and analyze flood forecasting models using machine learning algorithms within the Orange visual software environment. The study employed five algorithms: Random Forest, Decision Tree, Gradient Boosting, AdaBoost, and K-Nearest Neighbors. A comparative analysis of these models was conducted using key classification metrics, including Accuracy, Precision, Recall, and AUC. Special emphasis was placed on visualizing the results and assessing the usability of models within the Orange environment. The findings can be valuable for educational purposes, such as teaching students to work with real environmental data, as well as for practical applications in early warning systems and flood monitoring.