Symbolic Representation of Electromagnetic Data for Seismic Anomaly Detection

摘要

In this work we investigate the use of symbolicrnrepresentation methods for Anomaly Detection inrndifferent electromagnetic sequential time seriesrndatasets. An issue that is often overlooked regardingrnsymbolic representation and its performance inrnAnomaly Detection is the use of a quantitative accuracyrnmetric. Until recently only visual representations havernbeen used to show the efficiency of an algorithm torndetect anomalies. In this respect we propose an novelrnaccuracy metric that takes into account the length ofrnthe sliding window of such symbolic representationrnalgorithms and we present its utility. For the evaluationrnof the accuracy metric, HOT-SAX is used, a method thatrnaggregates data points by use of sliding windows. ArnHOT-SAX variant, with the use of overlapping windows,rnis also introduced that achieves better results based onrnthe newly defined accuracy metric. Both algorithms arernevaluated under ten benchmark and real terrestrial andrnsatellite data.