Spatial analisys of magnitude distribution for earthquake prediction using neural network based on automatic clustering in Indonesia

Abstract

A spatial analysis of magnitude distribution is presented in this paper to identify the optimal number of clusters based on seismic data of all region in Indonesia. The data were obtained from Indonesian Agency for Meteorological, Climatological and Geophysics (BMKG) and United States Geological Survey's (USGS). Clustering process consist of two steps: finding the global optimum number of clusters using Valley Tracing and clustering the dataset based on Hierarchical K-means. The optimal number of cluster obtained is 6 cluster. A model of Artificial Neural Networks (ANNs) is implemented for selected cluster to conduct an earthquake prediction. The architecture of the neural network model is composed of seven inputs, two hidden layers with thirty-two nodes each and one output. Back propagation training method and sigmoid activation function are applied. The input values are related to the b-value, the Bath's law, and the Omori-Utsu's law. The ANNs prototype predicts earthquake which is equal or larger than the given threshold magnitude during the next five days after an earthquake occurrence. Statistical tests are provided using two threshold values (5.5 and 6). The ANNs result showed that the proposed model gave better performance to predict earthquake that equal or larger than 6 Richter's scale magnitude. Finally, the result were compared to other ANNs model showing quantitatively and qualitatively better results.