Toward Location Reliant Early Earthquake Detection: A Paradigm Deployed Deep Learning Algorithms and Clustered Seismic Indicators

We present a deep learning approach for location-dependent earthquake prediction in seismically active regions using LSTM-based models. The Himalayan belt served as the test region, segmented into seven clusters via k-means, with earthquake data further divided by magnitude. Seventeen seismic indicators were used to train LSTM, Bi-LSTM, and Conv-LSTM models, apprehending spatiotemporal patterns. Results show that the LSTM model provides the highest accuracy, with MAE ranging 0.08–0.25 for time difference prediction and 0.016–0.11 for magnitude prediction. Including regional features improved predictions, making the method suitable for early warning systems, risk forecasting, and disaster mitigation in earthquake-prone areas.