Advancing Flood Prediction and Classification: A Systematic Literature Review

Abstract

Floods are the most frequent and destructive natural hazards, intensified by climate change and rapid urbanization, which create an urgent need for accurate prediction and classification techniques to mitigate the challenges. While machine learning (ML) and deep learning (DL) techniques have shown great promise, Unlike previous reviews that remain fragmented by examining only forecasting, prediction or susceptibility mapping, this study integrates prediction and classification tasks, systematically analyzes datasets and evaluation metrics, and develops a taxonomy linking data, methods, and performance by conducting a systematic literature review of 30 peer-reviewed articles published between the span of 2020 and 2024, retrieved from major academic databases, using explicit search strings and inclusion criteria. The synthesis reveals that traditional ML models such as Random Forests (RF), Support Vector Machines (SVM), and Artificial Neural Networks (ANN) remain effective for structured hydrological datasets, whereas DL models, particularly Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN), excel in capturing spatiotemporal flood dynamics. Beyond model performance, the review highlights critical backwards for research in Africa and the South American continent and methodological gaps, including data scarcity in underrepresented regions, inconsistent evaluation standards, limited adoption of explainable AI, and weak integration of socio-environmental variables. By providing a structured taxonomy that links datasets, methods, and performance metrics, this review advances beyond prior work and offers actionable insights for researchers and policymakers to strengthen flood risk modeling, disaster preparedness, and climate resilience.