Distinguishing Undesired and Desired Infrequent Behavior in Process Mining
MSc Thesis Project
Title: Distinguishing Undesired and Desired Infrequent Behavior in Process Mining
Author: Gyumin Lee
1st Examiner: Prof.dr.ir. Wil M.P. van der Aalst
2nd Examiner: Prof. Dr. rer. nat. Christoph Quix
Daily Supervisor: Mohammadreza Fani Sani
Process Mining techniques try to bridge the gap between business process science and classical data science. In Process Mining, users can obtain insights from the business pro- cess from the event log with Process Discovery. In Process Discovery, i.e., one of the main branches of Process Mining, many of Process Discovery algorithms are not able to handle undesired behaviors in the event log. In the real-life event log, undesired behaviors are spread in the event log. Moreover, current outlier filtering methods remove all infrequent behaviors instead of removing just undesired behaviors. Furthermore, in many applica- tions, we need to remove undesired behaviors but not just infrequent behaviors.
This thesis presents two filtering methods to distinguish undesired and desired in- frequent behavior. The first filtering method gives several options to let users detect and remove undesired behaviors. This work applied the mentioned method on two event logs that are labeled with an expert user, and users manually chooses undesired behaviors with their business knowledge by using an interactive user interface in the ProM framework. The second method exploits the data attributes such as waiting time and directly follows relation to obtaining the filtered event log. The evaluation has been evaluated through two labeled event logs obtained by the previous method. Also, the result shows that the proposed second method improves the quality of the event log and also the resulted process model.