Knowledge and Information Systems Human-Computer Interaction
Unifying global and local anomaly detection for time series
Abstract
Anomaly detection for time-series data has been proven useful from both academia and industry and can be categorized into global and local methods based on the range of data used to determine anomalies. To the best of our knowledge, this is the first study to systematically unify global and local anomaly detection methods for time-series data within a single framework. In this study, we propose an approach that unifies the global and local anomaly detection to achieve good performance. To make the unification possible, we discretize time-series data and partition the discrete data into global and local anomaly detection parts based on a concept of local factor. We design a rule-based global method using non-redundant association rules with tolerances to handle the difficulty of the latency phenomenon and the excessive number of rules in time-series data. We develop a Transformer-based local anomaly detection method with LSH (Locality Sensitive Hashing) Attention to better fit discrete data. We finally conduct an extensive experimental study to verify the effectiveness and efficiency of our approach. Compared with Anomaly Transformer (AT), our method achieves relative F1 improvements of 4.7%, 11.9%, and 7.0% on SWaT, WADI, and HAI, respectively, while reducing the corresponding running time by 43.0%, 47.0%, and 60.3%. Across the 48 TimeEval datasets, our method achieves a mean F1 score of 89.19%, compared with 84.34% for AT.