Anomaly Detection Using Multiscale Signatures

This paper analyzes multidimensional time series through the lens of their integrals of various moment orders, constituting their signatures, a novel tool for detecting anomalies in time series. The proposed anomaly detection (AD) method is compared using classical distance-based methods such as Local Outlier Factor (LOF) and One-Class Support Vector Machine (OCSVM). These methods are investigated using different similarity measures: distance on signature features, Euclidean distance and Dynamic Time Warping (DTW). The combination of signature features with a specific segmentation of time series leads to a multi-scale analysis tool that is competitive with respect to the state-of-the-art results, while maintaining low computational costs thanks to a property of the signature features.