Subset Multivariate Collective and Point Anomaly Detection

In the recent years, there has been a growing interest in identifying anomalous structure within multivariate data sequences. We consider problem of detecting collective anomalies, corresponding to intervals where one, or more, sequences behaves anomalously. first develop test for single anomaly that power simultaneously detect anomalies are either rare, is affecting few sequences, common. then show how multiple way computationally efficient but avoids approximations inherent binary segmentation-like approaches. This approach shown consistently estimate number and location anomalies—a property not previously competing methods. Our can be made robust point allow imperfectly aligned. practical usefulness allowing imperfect alignments through resulting increase regions copy variation. Supplemental files this article available online.