TU Darmstadt / ULB / TUbiblio

Detecting Concept Drift in Processes using Graph Metrics on Process Graphs

Seeliger, Alexander and Nolle, Timo and Mühlhäuser, Max
Mühlhäuser, Max and Zehbold, Cornelia (eds.) (2017):
Detecting Concept Drift in Processes using Graph Metrics on Process Graphs.
In: Proceedings of the 9th International Conference on Subject-oriented Business Process Management (S-BPM-ONE), New York, NY, USA, ACM, Darmstadt, Germany, In: S-BPM ONE '17, ISBN 978-1-4503-4862-1/17/03,
DOI: 10.1145/3040565.3040566, [Conference or Workshop Item]

Abstract

Work in organisations is often structured into business processes, implemented using process-aware information systems (PAISs). These systems aim to enforce employees to perform work in a certain way, executing tasks in a specified order. However, the execution strategy may change over time, leading to expected and unexpected changes in the overall process. Especially the unexpected changes may manifest without notice, which can have a big impact on the performance, costs, and compliance. Thus it is important to detect these hidden changes early in order to prevent monetary consequences. Traditional process mining techniques are unable to identify these execution changes because they usually generalise without considering time as an extra dimension, and assume stable processes. Most algorithms only produce a single process model, reflecting the behaviour of the complete analysis scope. Small changes cannot be identified as they only occur in a small part of the event log. This paper proposes a method to detect process drifts by performing statistical tests on graph metrics calculated from discovered process models. Using process models allows to additionally gather details about the structure of the drift to answer the question which changes were made to the process.

Item Type: Conference or Workshop Item
Erschienen: 2017
Editors: Mühlhäuser, Max and Zehbold, Cornelia
Creators: Seeliger, Alexander and Nolle, Timo and Mühlhäuser, Max
Title: Detecting Concept Drift in Processes using Graph Metrics on Process Graphs
Language: English
Abstract:

Work in organisations is often structured into business processes, implemented using process-aware information systems (PAISs). These systems aim to enforce employees to perform work in a certain way, executing tasks in a specified order. However, the execution strategy may change over time, leading to expected and unexpected changes in the overall process. Especially the unexpected changes may manifest without notice, which can have a big impact on the performance, costs, and compliance. Thus it is important to detect these hidden changes early in order to prevent monetary consequences. Traditional process mining techniques are unable to identify these execution changes because they usually generalise without considering time as an extra dimension, and assume stable processes. Most algorithms only produce a single process model, reflecting the behaviour of the complete analysis scope. Small changes cannot be identified as they only occur in a small part of the event log. This paper proposes a method to detect process drifts by performing statistical tests on graph metrics calculated from discovered process models. Using process models allows to additionally gather details about the structure of the drift to answer the question which changes were made to the process.

Title of Book: Proceedings of the 9th International Conference on Subject-oriented Business Process Management (S-BPM-ONE)
Series Name: S-BPM ONE '17
Volume: 9
Place of Publication: New York, NY, USA
Publisher: ACM
ISBN: 978-1-4503-4862-1/17/03
Uncontrolled Keywords: Change Point Detection; Concept Drift; Process Drift; Process Mining; Process Dynamics
Divisions: 20 Department of Computer Science
20 Department of Computer Science > Telecooperation
LOEWE
LOEWE > LOEWE-Schwerpunkte
LOEWE > LOEWE-Schwerpunkte > NiCER – Networked infrastructureless Cooperation for Emergency Response
Event Location: Darmstadt, Germany
Date Deposited: 19 Jan 2017 08:44
DOI: 10.1145/3040565.3040566
Identification Number: TUD-CS-2017-0023
Related URLs:
Export:

Optionen (nur für Redakteure)

View Item View Item