Earthquake Magnitude and Depth Prediction Based on Hybrid GRU-BiLSTM Model

An earthquake is a natural calamity and cannot be avoided or stopped, but early steps can be taken to minimize human casualties in addition to financial, infrastructural and environmental loss. Earthquake data is sequentially complex, which makes finding any relation in between previous earthquakes and future ones difficult. However, the development of deep learning-based neural networks shows a promising scope in this direction and significant research is being carried out to predict earthquakes even before occurrence. This paper is such an attempt, to develop a model which can find relation between time, location, magnitude and depth of an earthquake event based on previously available data. The proposed model is hybrid Gated Recurrent Unit Bidirectional Long Short-Term Memory Network (GRU-BiLSTM) model, trained on dataset containing information about earthquakes that have occurred in Japan from 1960 to 2019, with a magnitude of more than 4.0. Furthermore, the performance of proposed model is compared with other models which have been previously developed.