Seismic source modeling by clustering earthquakes and predicting earthquake magnitudes

Seismic sources are currently generated manually by experts, a process which is not efficient as the size of historical earthquake databases is growing. However, large historical earthquake databases provide an opportunity to generate seismic sources through data mining techniques. In this paper, we propose hierarchical clustering of historical earthquakes for generating seismic sources automatically. To evaluate the effectiveness of clustering in producing homogenous seismic sources, we compare the accuracy of earthquake magnitude prediction models before and after clustering. Three prediction models are experimented: decision tree, SVM, and kNN. The results show that: (1) the clustering approach leads to improved accuracy of prediction models; (2) the most accurate prediction model and the most homogenous seismic sources are achieved when earthquakes are clustered based on their non-spatial attributes; and (3) among the three prediction models experimented in this work, decision tree is the most accurate one.