Flooding in coastal regions of smart cities poses significant challenges, including infrastructure damage, economic losses, and threats to public safety. Traditional flood prediction models often suffer from data privacy concerns, limited spatial-temporal generalisation, and computational inefficiencies. To address these challenges, this study proposes an advanced Federated Learning (FL) and CNN-LSTM-based predictive framework for flood forecasting in coastal urban regions. The FL paradigm enables decentralised model training across multiple locations while ensuring data privacy. Convolutional Neural Networks (CNNs) extract spatial flood-related features, while Long Short-Term Memory (LSTM) networks capture temporal dependencies in hydrometeorological data. Various sensors, IoT devices and geospatial equipment are deployed to monitor and record flood-related environmental factors in different coastal regions in smart cities. The generated data is analysed by CNN and LSTM models to predict the flood levels based on the flood-influencing factors estimated. The proposed FL-CNN-LSTM model is implemented and experimented with in Python, and the prediction efficiency is verified. It is also compared with the other earlier methods and evaluates performance. It shows that the FL-CNN-LSTM provides more accuracy and promising quality services like dependency reduction in centralised data storage, adaptiveness, and privacy preservation in flood forecasting systems. Most importantly, it provides a proactive natural disaster mitigation model, making it suitable for real-time coastal regions in smart cities.
Total file downloads: 1