Identification and characterisation of earthquake clusters from seismic historical data

Abstract

New approaches and methods based on machine learning technologies make it possible to identify not only the spread of earthquakes, but also to establish hidden patterns that allow further assessment of any risks associated with their occurrence. In this article, the clustering algorithms of K-means and K-medoids are applied for the analysis of seismic data recorded on the territory of the Republic of Kazakhstan. Using the Elbow and Silhouette methods, the optimal value of K clusters was determined, which was later used in classifying a data set using cluster analysis methods. The results of seismic data classification by clustering algorithms are in line with expectations. However, when measuring the quality of clustering, the accuracy of the model by the K-means method exceeded the accuracy of the K-medoids model, and the scoring value by the K-means method is ahead of the value by the K-medoids method. In addition, the presented results of descriptive statistics allowed to carry out a more in-depth analysis of the characteristics of each cluster.