SEISMIC ANOMALY DETECTION USING SYMBOLIC REPRESENTATION METHODS

摘要

In this work we investigate the use of symbolic representationrnmethods for Anomaly Detection in different electromagneticrnsequential time series datasets. An issuernthat is often overlooked regarding symbolic representationrnand its performance in Anomaly Detection is the usernof a quantitative accuracy metric. Until recently only visualrnrepresentations have been used to show the efficiencyrnof an algorithm to detect anomalies. In this respect wernpropose an novel accuracy metric that takes into accountrnthe length of the sliding window of such symbolic representationrnalgorithms and we present its utility. For thernevaluation of the accuracy metric, HOT-SAX is used, arnmethod that aggregates data points by use of sliding windows.rnA HOT-SAX variant, with the use of overlappingrnwindows, is also introduced that achieves better resultsrnbased on the newly defined accuracy metric. Both methodsrnare evaluated on ten different benchmark datasets andrnbased on the empirical evidence we use Earth’s geomagneticrndata gathered by the SWARM satellites and terrestrialrnsources around the epicenter of two seismic eventsrnin the Yunnan region of China.