Discovery of Local Process Models by Combining Places
MSc Thesis Project
Title: Discovery of Local Process Models by Combining Places
Author: Viki Peeva
Supervisor: Lisa Mannel
1st examiner: Prof.Dr. ir. Wil M.P. van der Aalst
2nd examiner: Prof. Dr. Martin Grohe
In this thesis, we describe a new method for finding local process models by combining places. Process discovery algorithms can sometimes fail to find well-structured process models, particularly when there is high variability in the event log. For this reason, we turn ourselves to local process model discovery algorithms, which focus on finding local patterns in the event log. Our method finds local process models for some limited local distance by combining places given by an oracle. We use a sliding window to express the local distance, and we introduce two tree structures that we use to build the local process models. Using our structures, we can find all local process models by passing the event log only once, making our approach suitable for large event logs. Our algorithm need a lot of space when it uses many places or larger local distances. We aim to reduce this and at the same time return better quality local process models by introducing different evaluation metrics that measure the quality of the places we use and the local process models we return. In our evaluation, we analyse the correlation between different parameters and the running time on multiple event logs, confirming that the size of the event log is not critical. Additionally, we compare our approach with an existing one, and we see that if we have an appropriate set of places, we can also find the local process models that the other approach finds. Our algorithm can find substantial quantities of local process models, so we use the different evaluation metrics to rank our local process models.