Predquake: A Deep Learning Algorithm to Predict The Time and Location of Earthquakes

Earthquakes are the most dangerous natural disasters, causing significant damage globally. From 1998-2017, they caused a total of 750,000 deaths, over half of all deaths from natural disasters during that time. Moreover, they cause billions of dollars in damage annually, impacting many people. Seismologists work with other scientists to attempt to solve this problem by predicting earthquakes before they happen. However, despite many attempts, they are unsuccessful in predicting the time, location, and magnitude of a future earthquake. This research proposes a deep learning based algorithm, Predquake, to accurately predict the time and location of earthquakes internationally-two out of the three aspects of earthquake prediction. By combining the vast amounts of earthquake data gathered from seismometers across the world with advanced deep learning techniques, it presents high potential for success. The model uses a LSTM based regression approach across past earthquake information to make accurate predictions. The research finds that the model has a low accuracy for location prediction outside of the Southern Europe region, but a high accuracy for time prediction, being able to predict the next earthquake with 3 hours of accuracy. Although the algorithm is unable to predict exact locations of future earthquakes, it presents a significant step towards complete prediction. Being able to forecast the time of future quakes can help cities prepare for an upcoming disaster as well as help seismologists discover trends in earthquake occurrences.