Enhancing Earthquake Prediction and Early Warning Systems using CapsNet-BiLSTM Models

EEW are crucial for enhancing catastrophe resilience by delivering real-time notifications to individuals, governments, communities, and businesses. These devices assist in reducing damage and loss by facilitating timely intervention prior to the seismic activity impacting impacted regions. The efficacy of EEW relies on precise data processing, necessitating sophisticated methods to reduce mistakes induced by noise and data discrepancies. This research introduces a multi-phase earthquake prediction system that combines a 1D CapsNet with a LSTM network. Wavelet techniques are utilised for pre-processing to mitigate noise and outliers, hence improving data balance, while a hybrid Feature Processing Algorithm (FPA) is implemented for feature extraction. The LSTM network identifies temporal relationships in seismic data, whereas the 1D CapsNet extracts advanced features from sequential input, enhancing total prediction accuracy. The suggested 1D-Caps-LSTM model has exceptional efficacy in earthquake prediction, surpassing other methodologies with an accuracy of 95.64%. The results underscore the model's efficacy in augmenting EEW systems and bolstering disaster preparedness. The proposed architecture utilises deep learning techniques to provide a dependable mechanism for prompt earthquake alerts. Subsequent study may concentrate on real-time implementation and the expansion of the system to various geographical areas.