Abstract
Concept drifts within business processes are viewed as variations in the business circumstances, such as structural and behavioural changes in the control-flow, which necessitate process refinement and model updating. Existing approaches, such as relation-based precedence rules, tuned to detect drifts in the process structure are often not well suited to detecting changes in customer behaviour. This paper proposes a concept drift detector employing multi-components originating from Discrete-time Markov chains to detect, localize and reason about concept drifts in both process structure and customer behaviour of the control-flow. The approach was compared with three commonly used methods using 52 artificial event logs representing various types of drift (sudden and gradual, structural and behavioural). Experimental results demonstrated desirable performance with average F1 scores of 0.871 and 0.893 under structural and behavioural drifts, respectively. The approach was also employed in a real-life hospital billing dataset. The main contribution of this paper is a concept drift detector that is able to detect and explain root causes of control-flow changes whether such variations occurred suddenly or gradually.
Original language | English |
---|---|
Article number | 118533 |
Pages (from-to) | 1-14 |
Number of pages | 14 |
Journal | Expert Systems with Applications |
Volume | 210 |
Early online date | 17 Aug 2022 |
DOIs | |
Publication status | Published (in print/issue) - 30 Dec 2022 |
Bibliographical note
Funding Information:This research is supported by BTIIC (the BT Ireland Innovation Centre), funded by BT and Invest Northern Ireland . All authors have read and agreed to the published version of the manuscript.
Publisher Copyright:
© 2022 Elsevier Ltd
Keywords
- Business process
- Concept drift
- Behavioural drift
- Discrete-time Markov chains
- Sliding window