A Multi-feature Anomaly Detection Method Based on AETA ULF Electromagnetic Disturbance Signal

There have been many studies in relationship between ultra-low frequency electromagnetic anomaly and earthquakes, while most of them judge anomaly using single feature. We propose a multi-feature anomaly detection method for AETA ULF electromagnetic disturbance signals based on Isolation Forest, with some feature extraction and selection method added. A statistical test method superposed epoch analysis (SEA) is used for its evaluation. The result shows that 6 of 12 selected stations show significant correlation between signal anomaly and earthquakes. A further comparison experiment shows that our method has better performance than traditional single-feature sliding IQR method, which indicates multi-feature might be a good choice in finding global anomaly points.