Continual Anomaly Detection: A Comprehensive Survey and Research Roadmap

Continual Anomaly Detection (CAD) addresses the challenge of identifying abnormal patterns in evolving data by integrating adaptation to new conditions with knowledge retention. Since real-world applications deal with dynamic and non-stationary environments, anomaly detection models must continuously evolve to remain effective. In this context, CAD, sometimes also referred to as lifelong anomaly detection, has recently emerged as a distinct research area at the intersection of Continual Learning (CL) and Anomaly Detection (AD). This survey discusses: i) CAD learning scenarios and the assumptions they impose, ii) methods for adaptation and retention, iii) application domains, and iv) evaluation metrics that capture both detection quality and continual dynamics. We identify key gaps in current research, including limitations in current approaches, scenarios, evaluation, and alignment with real-world conditions. Finally, we discuss open challenges and propose a roadmap to guide future research toward robust anomaly detection systems suitable for evolving real-world environments.