Earthquake Prediction Based on Machine Learning Techniques

Abstract

Earthquakes are a natural calamity produced by the movement of the earth's tectonic plates due to enormous internal energy being released, earthquakes can cause in serious injuries and fatalities, demolish massive structures and infrastructure leading to significant economic loss. Earthquakes predictions are achieving society's safety, reduce the magnitude of destruction. For predicting an earthquake's time, magnitude, depth, and location, a variety of techniques have been suggested, such as statistical and mathematical analysis and a signal investigation of precursors and due to an ostensibly dynamic character of seismic, they usually do not produce excellent results. The ability of artificial intelligence to detect hidden patterns of data and nonlinear relation well-known has been gaining attention in recent years. It has been used in several areas and achieved positive results. This thesis utilizes artificial intelligence algorithms in predicting the next earthquakes to take the necessary precautions and reduce the risk in earthquake-prone areas. The support vector regression, feed-forward neural network and long short-term memory algorithm were applied to predict the occurrence of the next earthquake based on the historical data obtained from the General Directorate of Meteorology and Earthquake Monitoring in Iraq, through study data for three different regions in Iraq (Sulaymaniyah, Maysan and Wasit) to predict earthquake characteristics (time, magnitude, location and depth) and evaluating the performance of the prediction using testing data The obtained results are compared by evaluation metrics to show which algorithm is the best for earthquake prediction. From these results, it is concluded that the long short term memory is the best for earthquake prediction and assists in obtaining promising results, because Long Short Term Memory is an optimized neural network that can handle issues of exploding and vanishing gradients. This algorithm where achieved 87% accuracy in predicting the time and magnitude of the earthquake, 88% accuracy in predicting time, magnitude, and location of the earthquake, and 91% accuracy in predicting the magnitude, location, time and depth of the earthquake