Identifying Devastated Flood Events using CNN Algorithm

Abstract

Tsunami prediction requires accurate analysis of seismic and oceanographic parameters to support timely disaster mitigation. This study applies a Convolutional Neural Network (CNN) to classify tsunami-related and non-tsunami events using historical data from the National Oceanic and Atmospheric Administration (NOAA) tsunami dataset. We utilized relevant features, including earthquake magnitude, depth, location, wave height, and other tsunami-related parameters, to train and evaluate the model. The CNN learned nonlinear patterns among these variables and achieved 92% accuracy on the test dataset. The evaluation also produced 90% precision, 94% recall, and 92% F1-score, demonstrating consistent classification performance. The high recall indicates that the model successfully identified most actual tsunami events, which is particularly important in disaster prediction applications where missed tsunami events may lead to serious consequences. We also applied a probability threshold of 0.5 to determine the predicted class, and the prediction results showed that most samples obtained probabilities clearly above or below the threshold. These findings indicate that the proposed CNN can effectively learn patterns from historical seismic and oceanographic data.