An Earthquake Prediction System for Bangladesh Using Deep Long Short-Term Memory Architecture

Earthquake is a natural catastrophe, which is one of the most significant causes of structural and financial damage, along with the death of many humans. Prediction of the earthquake at least a month ahead of the event may diminish the death toll and financial loss. Bangladesh is in an active seismic region, where many earthquakes with small and medium magnitude occur almost every year. Several scientists have predicted that there is a good chance of an earthquake with startling energy shortly in this region. In this work, we have proposed a long short-term memory (LSTM)-based architecture for earthquake prediction in Bangladesh in the following month. After tuning hyperparameters, an architecture of 2 LSTM layers with 200 and 100 neurons, respectively, along with L1 and L2 regularization, was found to be the most efficient. The activation functions of the LSTM layers were tanh in the proposed architecture. The proposed LSTM architecture achieved a remarkable 70.67% accuracy with 64.78% sensitivity, 75.94% specificity in earthquake prediction for this region.