Natural Disaster Prediction Using Deep Learning

Abstract

Accurate prediction techniques are essential in an era when natural disasters pose severe dangers to human lives and infrastructure. This article presents a deep learning approach that uses convolutional neural networks (CNNs) to anticipate natural disasters such as floods, earthquakes, wildfires, and storms. The work preprocesses a dataset of 4,429 pictures using scaling, normalization, and augmentation techniques (such as flipping and rotation) to ensure model resilience and diversity in the data. The image sets were separated into training, validation, and test sets to optimize model performance. Several CNN architectures were used to train the model, including Xception, VGG16, and MobileNet. The testing results showed that MobileNet outperformed the other models with a test accuracy of 92.25%. VGG16 finished second with 91.50%, while Xception achieved 91.00%. A thorough performance assessment was produced by examining these models according to precision, recall, and F1 score. The impressive outcomes, especially for MobileNet, are ascribed to its capacity to effectively extract salient characteristics from the pictures, like distinct patterns and textures connected to different kinds of calamities. The study's overall findings demonstrate that CNN-based models can make precise predictions about natural disasters in real-time situations. To improve early warning systems, the models in this work indicate a high degree of accuracy by merging image data from satellite sources and using advanced data augmentation techniques. The results show how deep learning can practically enhance emergency preparedness and response procedures.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 5 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.