Classification of Tectonic and Nontectonic Earthquakes by an Integrated Learning Algorithm

Abstract

A machine learning model for accurate and quick classification of tectonic and nontectonic earthquakes is proposed. Firstly, an improved method is proposed for detection of first arrivals. An iterative approach is applied for the multiwindow algorithm to decrease its computational cost, then a new method for detection of first arrivals is proposed by combining the improved multiwindow algorithm with a recursive least-squares filter and the Akaike information criterion. Secondly, it is shown that the integrated learning algorithm is a suitable method for classification, then four suitable features are manually selected to train it. Thirdly, the three parameter values of the integrated learning model are determined to improve its accuracy. Finally, based on seismic data from the China Earthquake Networks Center, simulations are conducted to test the validity of the proposed method for detection of first arrivals, then the classification accuracy is tested. The results show an accuracy of the model of 88.88%, indicating effective classification performance.