Floods are among the most destructive natural dis asters, causing substantial losses worldwide. Advances in machine learning (ML) have introduced innovative approaches for flood prediction, offering enhanced accuracy and timeliness compared to traditional methods. This paper presents a comprehensive review of 15 studies employing various ML algorithms, including Decision Trees, Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Deep Learning, in flood prediction systems. The review provides a detailed comparison of models, emphasizing performance metrics such as accuracy, precision, and computational efficiency. Unique contributions include the identification of key research gaps such as the absence of real-time adaptability, IoT integration, and the limited use of state-of-the-art models like transformers and the proposal of hybrid and ensemble modeling strategies for improved resilience and predictive accuracy. Practical case studies and applications of these methods are discussed to highlight their feasibility and impact. The study also emphasizes the integration of IoT data, hybrid models, advanced spatial analysis, and transformer based architectures to enhance global flood forecasting systems, bridging current gaps and paving the way for more robust and adaptive predictive frameworks.
Dhouni et al. (Wed,) studied this question.