Section: New Results

Anomaly Detection in Streams with Extreme Value Theory

Participants : Alban Siffer [EMSEC] , Pierre-Alain Fouque [EMSEC] , Christine Largouët, Alexandre Termier.

Anomaly detection in time series has attracted considerable attention due to its importance in many real-world applications including intrusion detection, energy management and finance. Most approaches for detecting outliers rely on either manually set thresholds or assumptions on the distribution of data. In [16], we propose a new approach to detect outliers in streaming univariate time series based on Extreme Value Theory that does not require to handpick thresholds and makes no assumption on the distribution: the main parameter is only the risk, controlling the number of false positives. Our approach can be used for outlier detection, but more generally for automatically setting thresholds, making it useful in wide number of situations. We also test our algorithms on various real-world datasets which confirm the soundness and efficiency of our methods.