An unsupervised anomaly detection approach for pre-seismic ionospheric total electron content

This paper focuses on the anomaly detection for ionospheric total electron content (TEC) before earthquakes. In this paper, a novel unsupervised approach is proposed. First, interval-based method is employed to granulate the TEC series. Justifiable granularity principle is utilized to construct interval information granules (IGs) for representing TEC series. Second, high-order difference method is introduced to construct rectangle IGs and cube IGs for obtaining the new representation of TEC. Third, corresponding similarity measurement method is designed to calculate the anomaly score of each IG, which is the evaluation criterion for detecting the anomalies. Finally, experimental results using real TEC datasets validate the effectiveness of the proposed approach. Compared with the existing major approaches, because the proposed approach can capture more morphological details and variation trend of TEC series, it can achieve a higher detection accuracy.