In this work we investigate the use of symbolic represen- tation methods for Anomaly Detection in different elec- tromagnetic sequential time series datasets. An issue that is often overlooked regarding symbolic representa- tion and its performance in Anomaly Detection is the use of a quantitative accuracy metric. Until recently only vi- sual representations have been used to show the efficiency of an algorithm to detect anomalies. In this respect we propose an novel accuracy metric that takes into account the length of the sliding window of such symbolic rep- resentation algorithms and we present its utility. For the evaluation of the accuracy metric, HOT-SAX is used, a method that aggregates data points by use of sliding win- dows. A HOT-SAX variant, with the use of overlapping windows, is also introduced that achieves better results based on the newly defined accuracy metric. Both meth- ods are evaluated on ten different benchmark datasets and the Earth’s geomagnetic data gathered by the SWARM satellites and terrestrial sources around the epicenter of two seismic events in the Yunnan region of China.
|Title of host publication||Unknown Host Publication|
|Number of pages||8|
|Publication status||Published - 12 Sep 2016|
|Event||Dragon 3 symposium - |
Duration: 12 Sep 2016 → …
|Conference||Dragon 3 symposium|
|Period||12/09/16 → …|
- Seismic Anomaly Detection
- Symbolic Rep- resentation of Time Series Data and Accuracy Measure.
Christodoulou, V., Bi, Y., Wilkie, G., & Zhao, G. (2016). Seismic Anomaly Detection Using Symbolic Representation Methods. In Unknown Host Publication ESA Communications. http://uir.ulster.ac.uk/36235/2/Accept.pdf