Please use this identifier to cite or link to this item:
http://hdl.handle.net/1942/28268
Title: | Unearthing the Real Process Behind the Event Data: The Case for Increased Process Realism | Authors: | JANSSENSWILLEN, Gert | Advisors: | DEPAIRE, Benoit | Issue Date: | 2019 | Abstract: | Companies in the 21st century posses a large amount of data about their products, customers and transactions. The prominent role of business processes in the modern organisation in recent decades has led to a remarkable increase in the amount of event data that is available. Event logs are logbooks that contain information about everything that happens in a company on a daily basis. A customer who places an order, an employee who logs in to the customer management system to handle the order, a supplier who delivers a quotation for the products, a production line that is started, etc. The digitisation of all these events enables us to analyse business processes at a level that was previously unimaginable. The increase in available event data gave rise to process mining, a discipline that focuses on extracting insights about processes from event logs. However, correctly displaying business processes is not a trivial task. Due to the high complexity of most processes, event logs contain only a limited sample of all the possible ways and combinations in which business processes can be performed. Errors and inconsistencies in the available data create additional difficulties. In response to these challenges, process discovery algorithms were developed - algorithms that discover process models based on event logs. However, the crucial question is: how good are these discovered models? Are they able to correctly represent business operations? The concept of process realism is introduced in this dissertation. To optimise processes, evidence-based decision making is needed. Consequently, it is essential to map these processes in a realistic way. Blindly relying on both partial and / or inconsistent data and on algorithms can lead to wrong actions being taken. Process realism is approached from two perspectives in this dissertation. First, quality dimensions and measures for process discovery are analysed on a large scale and compared with each other on the basis of empirical experiments. Which measures are best suited to assess the quality of a discovered process model? What are their weaknesses and strengths? And what challenges still need to be overcome in order to evolve towards a reliable quality measurement? The experiments in this thesis show that there are important differences between the different quality measures in terms of feasibility, validity and sensitivity. Moreover, the role and meaning of the generalisation dimension is unclear. Existing generalisation measures do not succeed in adequately assessing the fit between process models and the underlying process. Fitness and precision measures also do not constitute unbiased estimators of the quality of the model as a representation of the underlying process. Furthermore, with regard to experimental set-up, various challenges have been identified that are necessary to evolve towards a correct quality measurement. In addition to the focus on process models, process realism is also approached from a data point of view. By developing a transparent and extensible tool-set, a framework is offered to analyse process data from different perspectives. Exploratory and descriptive analysis of process data and testing of hypotheses again leads to increased process realism. The developed framework is applied in this dissertation to two case studies. First, how can we use process data to better understand students’ study trajectories and to better guide students? Secondly, how can applying process analysis in a railway context to map out the use of the rail infrastructure and analyse deviations between the timetable and execution in order to achieve a smoother service for passengers? Both case studies show that the framework has clear added value, and that the answers to the questions asked can help to improve the processes under consideration. At the same time, however, unresolved challenges within process mining are also emphasised, such as the analysis of processes at the right level of granularity, and the assumption that process instances are independent of each other. From both perspectives, process models and process data, recommendations are made for future research, and a call is made to give the process realism mindset a central place within process mining analyses. | Document URI: | http://hdl.handle.net/1942/28268 | Category: | T1 | Type: | Theses and Dissertations |
Appears in Collections: | PhD theses Research publications |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
20190506 DOCTORAAT DEFINITIEF GERT JANSSENSWILLEN GJW_PHD_FINAL V2 correctie PAG346.pdf | 8.96 MB | Adobe PDF | View/Open |
Page view(s)
2,080
checked on Sep 7, 2022
Download(s)
78
checked on Sep 7, 2022
Google ScholarTM
Check
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.