Journal of Engineering Research and Reports

Aim: To develop an AI-powered real-time earthquake early warning system with deep learning techniques to enhance disaster preparedness and response by providing timely alerts and reducing the impact on human life and infrastructure.

Study Design: This was a model development and performance evaluation study based on deep learning for image classification.

Place and Duration of Study: The study was conducted at the Department of Information Technology, [Dhanalakshmi Srinivasan Engineering College (Autonomous)], over six months [December 2024] to [May 2025].

Methodology: The methodology for an AI-powered real-time earthquake early warning system using deep learning begins with collecting and preprocessing seismic data from sources like the USGS and local sensors. This data is cleaned, normalized, and converted into formats suitable for model training. Features such as wave arrival times and amplitudes are extracted, and deep learning models like CNNs, RNNs,and Multilayer perceptron (MLPs) are trained to detect and classify seismic events. These models are deployed in a real-time pipeline to analyze live data and trigger alerts based on predefined thresholds. The system is tested for accuracy and continuously updated with new data to improve reliability and responsiveness.

Results: The AI-powered earthquake early warning system demonstrated strong performance, achieving 98.2% accuracy on the validation dataset. High precision ensured relevant alerts with minimal false alarms, while high recall indicated effective detection of most seismic events, including early tremors. The F1-score confirmed a balanced performance, especially valuable for imbalanced data. Additionally, the system exhibited low latency, enabling real-time detection and timely alerts before damaging shockwaves arrive.

Conclusion: This research highlights the potential of Convolutional Neural Networks (CNNs) and ResNet algorithms for earthquake early warning (EEW) systems in India. The CNN-based model effectively distinguishes between earthquake and non-earthquake signals with high accuracy and low latency, enabling timely alerts and faster responses. Similarly, the ResNet algorithm, with its deep architecture and residual connections, enhances the model’s ability to learn complex seismic patterns and achieve even higher accuracy.