Improving Anomaly Detection with Adaptive Dynamic Threshold: A Review and Enhanced Method
Arman Aghaei Attar, Kaibin Bao, Veit Hagenmeyer, Tagir Fabarisov, Andrey S. Morozov · 2024
In fast-growing modern cyber-physical systems, reliability plays a vital role. Effective anomaly detection, which identifies security and safety issues at early stage, is essential for ensuring system reliability. While much research has focused on anomaly detection techniques, fewer studies address a key challenge, setting precise and responsive thresholds for anomaly detection. In this study, we conduct a comprehensive review of current threshold setting methods. Thereafter, we introduce a novel approach for adaptive threshold setting. Our method is tailored for diverse safety and security tasks and is tested on a safety-critical exoskeleton model and a cybersecurity scenario for energy systems. The results demonstrate that our approach could enhance the threshold setting for anomaly detection in CPS.