Classification of geomagnetic variation related to earthquakes using supervised learning algorithms

The various scientific techniques have been studied in order to predict the geological hazards such as earthquake, volcano, landslide. Particularly, in the case of earthquakes, the study using geomagnetic field has been focused to understand the precursors (Fraser-Smith et al., 1990; Hayakawa et al., 1996; Kawate et al., 1998; Hattori et al., 1998; Oh, 2012). This study is based on Oh's method (2012) and applied to supervised learning algorithms so as to classify for geomagnetic variation induced by earthquake. Main concept of Oh's method consists of data reconstruction by Principal Component Analysis and its application to wavelet based semblance method. These techniques show semblance anomaly between observation data and reconstructed data for specific events, and we have estimated the anomaly caused by earthquake and have been verified for earthquake case in Korea. In this study, we used the results not for a specific case but many cases of earthquake as supervised learning data. Supervised learning creates classification model using simple information such as coordinate of epicenter, magnitude, and depth with semblance result index. The generated model is used if the geomagnetic variation can be predicted when only the simple earthquake information is provided excluding semblance index. This results could be used to study the distance relationship between epicenter and geomagnetic observatories, and which range of magnitude may have an effect on geomagnetic fields.