Adaptive Dynamic Thresholds for Unsupervised Joint Anomaly Detection and Trend Prediction

Anomaly detection and degradation trend prediction are two pivotal tasks in system health management. However, most existing approaches treat them as independent problems and fail to exploit their intrinsic interdependence. In addition, the scarcity of labeled data in real-world scenarios limits the applicability of supervised learning methods. To address these challenges, we propose an adaptive thresholding strategy framework for unsupervised joint anomaly detection and trend prediction. Our framework introduces a self-adaptive threshold strategy from historical data distributions and dynamically updates them in response to evolving system behavior. The anomaly detection results are integrated to enhance degradation trend forecasting, while the predicted degradation trends, in turn, refine the anomaly thresholds through a feedback mechanism. Experiments on both public and real-world industrial datasets demonstrate that the proposed framework achieves superior detection accuracy, robust trend prediction, and high computational efficiency under diverse operational conditions.