TU Darmstadt / ULB / TUbiblio

BINet: Multivariate Business Process Anomaly Detection Using Deep Learning

Nolle, Timo and Seeliger, Alexander and Mühlhäuser, Max
Weske, Mathias and Montali, Marco and Weber, Ingo and vom Brocke, Jan (eds.) (2018):
BINet: Multivariate Business Process Anomaly Detection Using Deep Learning.
Springer, Sydney, Australia, September 9-14, 2018, ISBN 978-3-319-98647-0,
DOI: 10.1007/978-3-319-98648-7_16, [Online-Edition: https://doi.org/10.1007/978-3-319-98648-7_16],
[Conference or Workshop Item]

Abstract

In this paper, we propose BINet, a neural network architecture for real-time multivariate anomaly detection in business process event logs. BINet has been designed to handle both the control flow and the data perspective of a business process. Additionally, we propose a heuristic for setting the threshold of an anomaly detection algorithm automatically. We demonstrate that BINet can be used to detect anomalies in event logs not only on a case level, but also on event attribute level. We compare BINet to 6 other state-of-the-art anomaly detection algorithms and evaluate their performance on an elaborate data corpus of 60 synthetic and 21 real life event logs using artificial anomalies. BINet reached an average F₁ score over all detection levels of 0.83, whereas the next best approach, a denoising autoencoder, reached only 0.74. This F₁ score is calculated over two different levels of detection, namely case and attribute level. BINet reached 0.84 on case and 0.82 on attribute level, whereas the next best approach reached 0.78 and 0.71 respectively.

Item Type: Conference or Workshop Item
Erschienen: 2018
Editors: Weske, Mathias and Montali, Marco and Weber, Ingo and vom Brocke, Jan
Creators: Nolle, Timo and Seeliger, Alexander and Mühlhäuser, Max
Title: BINet: Multivariate Business Process Anomaly Detection Using Deep Learning
Language: English
Abstract:

In this paper, we propose BINet, a neural network architecture for real-time multivariate anomaly detection in business process event logs. BINet has been designed to handle both the control flow and the data perspective of a business process. Additionally, we propose a heuristic for setting the threshold of an anomaly detection algorithm automatically. We demonstrate that BINet can be used to detect anomalies in event logs not only on a case level, but also on event attribute level. We compare BINet to 6 other state-of-the-art anomaly detection algorithms and evaluate their performance on an elaborate data corpus of 60 synthetic and 21 real life event logs using artificial anomalies. BINet reached an average F₁ score over all detection levels of 0.83, whereas the next best approach, a denoising autoencoder, reached only 0.74. This F₁ score is calculated over two different levels of detection, namely case and attribute level. BINet reached 0.84 on case and 0.82 on attribute level, whereas the next best approach reached 0.78 and 0.71 respectively.

Journal or Publication Title: Business Process Management. BPM 2018. Lecture Notes in Computer Science
Publisher: Springer
Edition: 11080
ISBN: 978-3-319-98647-0
Divisions: 20 Department of Computer Science
20 Department of Computer Science > Telecooperation
Event Location: Sydney, Australia
Event Dates: September 9-14, 2018
Date Deposited: 29 Aug 2018 07:08
DOI: 10.1007/978-3-319-98648-7_16
Official URL: https://doi.org/10.1007/978-3-319-98648-7_16
Export:

Optionen (nur für Redakteure)

View Item View Item