Machine Learning-based Early Earthquake Magnitude Prediction using P-Wave Signal Feature Analysis

Early detection of earthquakes helps minimize the destructive impact of earthquakes and saves lives. Developing accurate and fast estimates of the size of earthquakes is extremely important since quick estimates can reduce damage from subsequent seismic activity. Recent developments in machine learning techniques appear to show a lot of promise for improving the performance of early warning systems in terms of their ability to produce accurate and fast estimates of earthquake size. This research will examine the application of machine learning techniques for predicting earthquakes prior to their arrival through an analysis of the characteristics of primary seismic waves (P-waves), which are the first seismic signal detected by seismometers when an earthquake first starts. Since P-waves travel faster than other seismic waves and arrive at a detector prior to the arrival of the more damaging secondary seismic waves, they provide a short but useful time interval for estimating earthquake size and generating an early warning to people within a certain distance of the earthquake. The various machine learning algorithms used to produce predictive models of earthquake size in this study include Random Forests, XGBoost, Support Vector Regressors, Decision Trees, K-Nearest Neighbors (KNNs), and ensemble-based machine learning techniques. The research used hundreds of thousands of records from a global earthquake database covering earthquakes occurring worldwide from 1995-2023 for model training and evaluation.