Abstract : This paper discusses several issues of evaluation and comparison of anomaly detection algorithms, namely lack of publicly available implementations and annotated data sets. Another problem of many methods is a detection delay caused by operating on data binned to a long time intervals. The paper presents a library under development which aims to tackle the comparison and evaluation issues. Further, the paper proposes a novel anomaly detection approach that can contribute to anomaly detection in real-time.
https://hal.inria.fr/hal-01529795
Contributor : Hal Ifip <>
Submitted on : Wednesday, May 31, 2017 - 1:17:54 PM Last modification on : Saturday, June 1, 2019 - 11:34:02 AM Long-term archiving on: : Wednesday, September 6, 2017 - 4:18:11 PM
Václav Bartoš, Martin Žádník. Network Anomaly Detection: Comparison and Real-Time Issues. 6th International Conference on Autonomous Infrastructure (AIMS), Jun 2012, Luxembourg, Luxembourg. pp.118-121, ⟨10.1007/978-3-642-30633-4_15⟩. ⟨hal-01529795⟩