Outlier/Anomaly Detection of Univariate Time Series: A Dataset Collection and Benchmark

David Muhr, Michael Affenzeller

Abstract: In this paper, we present an extensive collection of outlier/anomaly detection tasks to identify unusual series from a given time series dataset. The presented work is based on the popular UCR time series classification archive. In addition to the detection tasks, we provide curated benchmarks, an evaluation scheme and baseline results. The resulting unusual time series detection collection is openly available at: https://outlier-detection.github.io/utsd/ .