Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Filtering toolkit: Interactively filter event logs to improve the quality of discovered models

: Sani, M.F.; Berti, A.; Zelst, S.J. van; Aalst, W. van der

Volltext ()

Depaire, B.:
BPMT 2019. BPM 2019 Dissertation Award, Doctoral Consortium, and Demonstration Track. Online resource : Proceedings of the Dissertation Award, Doctoral Consortium, and Demonstration Track at BPM 2019 co-located with 17th International Conference on Business Process Management (BPM 2019), Vienna, Austria, September 1-6, 2019
La Clusaz: CEUR, 2019 (CEUR Workshop Proceedings 2420)
ISSN: 1613-0073
International Conference on Business Process Management (BPM) <17, 2019, Vienna>
Konferenzbeitrag, Elektronische Publikation
Fraunhofer FIT ()

Process discovery algorithms discover process models on the basis of
event data automatically. These techniques tend to consider the entire log to discover a process model. However, real-life event logs usually contain outlier behaviour that lead to incomprehensible, complex and inaccurate process models where correct and/or important behaviour is undetectable. Hence, removing outlier behaviour thanks to filtering techniques is an essential step to retrieve a good quality process model. Manually filtering the event log is tricky and requires a significant amount of time. On the other hand, some work in the past is focused on providing a fully automatic choice of the parameters of the discovery and filtering algorithms; however, the attempts were not completely successful. This demo paper describes an easy-to-use plug-in in the ProM process mining framework,
that provides a view where several process discovery and outlier filtering
algorithms can be chosen, along with their parameters, in order to find a sweet spot leading to a ’good’ process model. The filtered log is easily accessible, and the process model is shown inside the view, in this way the user can immediately evaluate the quality of the chosen combination between process discovery and filtering algorithms, and is effectively assisted in the choice of the preprocessing methodology. Some commonly used metrics (fitness, precision) are reported in the view provided by the plug-in, in order to ease the evaluation of the process model. With the options provided by our plug-in, the difficulties of both fullymanual and automatic choice of the filtering approach are effectively overcome.