FloodCNN-BiLSTM: Predicting flood events in urban environments

Abstract

A disaster is a severe event that occurs on a short period but has highly damaging and long-lasting effects on society. Disasters can be broadly categorized into natural and man-made events. Among natural disasters, floods are some of the most common disaster. As climate change accelerates, floods are expected to become more frequent and severe, highlighting the need for a deeper understanding of their causes, effects, and response strategies. Modern technologies, including machine learning, are increasingly being used to predict the occurrence of floods. Accurate forecasting requires large volumes of data collected from sensors deployed in various locations. Machine Learning (ML) models are well-suited for flood prediction due to their ability to handle sequential data and long-term dependencies. In this paper we present a hybrid deep learning model that combines Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks. The CNN component is responsible for extracting spatial features, while the BiLSTM processes the sequential data to classify the likelihood of flood events based on environmental parameters. The proposed FloodCNN-BiLSTM model has been validated on multiple datasets, achieving superior performance compared to traditional machine learning approaches. It attained 97.3 % accuracy on Dataset 1 and 98.6 % on Dataset 2. Evaluation metrics such as accuracy, precision, recall, and F1-score confirm the robustness and effectiveness of the model. Comparative analysis with other models used in this research demonstrates the superiority of our proposed approach.