An improved K-means clustering algorithm for global earthquake catalogs and earthquake magnitude prediction

Abstract

The occurrences of earthquakes have no any explicit warning, and earthquake magnitude prediction is still extremely challenging now. Therefore, this study proposes a seismic prediction model based on clustering of global earthquake data. First, an improved K-means clustering algorithm for global earthquake catalogs is proposed. Traditional K-means clustering has several limitations, i.e., the number of clusters needs to be initialized, the initial cluster centers are arbitrarily selected, and there is currently no magnitude parameter in the K-means clustering algorithm. To improve the algorithm, this study employs the space–time–magnitude (STM) distance and then proposes a maximum–minimum STM distance for the selection of the initial cluster centers. Additionally, the sum of squares error, Davies–Bouldin index, Calinski–Harabasz index, and silhouette coefficient are applied to determine the number of clusters. Subsequently, a seismic prediction model based on the clustering result combined with an artificial neural network is presented. Application of the improved clustering algorithm to the global seismic catalog from 1900–2019 obtained from the United States Geological Survey reveals better clustering accuracy than the traditional K-means algorithm and is also effective for seismic risk analysis in the local region. Furthermore, the seismic prediction model based on the clustering result also exhibits good performance, which has practical significance and reference value for future predictions of earthquake magnitude.