Loading [a11y]/accessibility-menu.js
Unsupervised multi scale anomaly detection in streams of events | IEEE Conference Publication | IEEE Xplore

Unsupervised multi scale anomaly detection in streams of events


Abstract:

Automatically detecting anomalies in streams of events is crucial for many applications in communications, security, healthcare, finance and real-time systems. In communi...Show More

Abstract:

Automatically detecting anomalies in streams of events is crucial for many applications in communications, security, healthcare, finance and real-time systems. In communication systems, it can be used to forecast equipment breakdowns or to detect unprecedented issues that do not trigger any alarms. Several methods have been proposed to detect anomalies in streams of events but they are not suited to detect large-scale anomalies with different durations and features. In this paper, we first propose a new data structure called s-digest to learn the distributions of values originating from streams of events for multiple time-scales. The structure is then used to conceive an unsupervised multi-scale method able to detect anomalies with different durations and characteristics. The method withstands high-throughput streams of events, is highly scalable and memory efficient. We then simulate a mobile network based on actual data from a commercial LTE network and apply our method to detect various anomalies and prove its accuracy and practicability.
Date of Conference: 19-21 December 2016
Date Added to IEEE Xplore: 06 February 2017
ISBN Information:
Conference Location: Surfers Paradise, QLD, Australia

References

References is not available for this document.