Online Conformal Anomaly Detection with Prediction-Powered Data Acquisition

Online anomaly detection is essential in fields such as cybersecurity, healthcare, industrial monitoring, and telecommunications, where promptly identifying deviations from expected behavior can avert critical failures or security breaches. While numerous anomaly scoring methods based on supervised or unsupervised learning have been proposed, the only existing approach capable of providing assumption-free guarantees on the false discovery rate (FDR) rely on a continuous stream of real-world calibration data. To address this limitation, we introduce context-aware prediction-powered conformal online anomaly detection (C-PP-COAD), a novel principled framework that strategically leverages synthetic calibration data to mitigate data scarcity, while adaptively integrating real data based on contextual information. C-PP-COAD wraps around any existing anomaly detection method, leveraging any given anomaly score to construct active conformal p-value statistics. These statistics support online testing with formal FDR control, maintaining rigorous and reliable anomaly detection performance over time. Experiments conducted on both synthetic and real-world datasets, including thyroid dysfunction detection, O-RAN conflict detection, 5G network intrusion detection, and O-RAN UE throughput degradation detection, demonstrate that C-PP-COAD significantly reduces dependency on real calibration data without compromising guaranteed FDR control.