A Systematic Literature Review on Classification Machine Learning for Urban Flood Hazard Mapping

The computational expensiveness of the hydrodynamic models and the complexity of the rainfall-runoff transformation process presents a pressing need to shift to machine learning (ML) for urban flood hazard mapping (UFHM). ML models, particularly classification techniques, have proven effective in complementing or replacing traditional methods. This paper systematically reviews the current state-of-the-art classification ML models, discussing the latest research papers published between 2018 and 2023 in which ML classification models are employed for UFHM. This systematic literature review (SLR) collected papers from five major digital libraries. Following a rigorous selection process, 81 articles were retained to address the research questions raised in this review. The SLR reveals that ensemble classifiers, followed by decision trees and artificial neural networks, demonstrate the best performance, whereas naive Bayes and k-nearest neighbors are less effective. Metaheuristic methods were the most frequently used for hyperparameter tuning. The majority of the selected studies primarily focus on pluvial and fluvial floods at regional and city scales. Our SLR also explores aspects such as the accuracy, the data preprocessing, and the input features of the ML models. The novelty of our review stems from being the first paper that systematically explores the classification ML models in urban flood hazard mapping. Additionally, recommendations on ML applications in urban flood mapping are provided.