A machine learning approach for asperities’ location identification

Abstract

Asperities’ location is a very important factor in spatiotemporal analysis of an area’s seismicity, as they can accumulate a large amount of tectonic stress and, by their rupture, a great magnitude earthquake. Seismic attributes of earthquakes, such as the b-value and seismic density, have been shown to be useful indicators of asperities’ location. In this work, machine learning techniques are used to identify the location of areas with high probability of asperity existence using as feature vector information extracted solely by earthquake catalogs (b-value and seismic density), avoiding thus any geo-location information. Extensive experimentation on algorithms’ performance is conducted with a plethora of machine learning classification algorithms, focusing on the effect of data oversampling and undersampling, as well as the effect of cost sensitive classification without any resampling of the data. The results obtained are promising with performance being comparable to geo-location information including vectors.