Improving the eST-Miner Models using Non-Unique Transition Labels



Lisa Mannel

Ph.D. Student


+49 241 80 21949



Improving the eST-Miner Models using Non-Unique Transition Labels

Student: Christian Rennert

Title: Improving the eST-Miner Models using Non-Unique Transition Labels

Supervisor: Lisa Mannel

1st Examiner: Prof. Wil M.P. van der Aalst

2nd Examiner: Prof. Martin Grohe


In this thesis, we introduce a new discovery algorithm in the field of process mining for the discovery of accepting Petri nets. Process discovery algorithms can discover imprecise structures like one-looping transitions based on the class of process models the discovery algorithm is restricted to or as a result of the structures that the discovery algorithm is limited to find. In this work, the eST-Miner is considered as discovery algorithm which is restricted to discover uniquely labeled Petri nets and therefore the eST-Miner is often discovering one-looping transitions. Here, our approach can be used to find a more precise non-uniquely labeled Petri net while preserving desirable structures existing in the input uniquely labeled Petri net. Further, we describe a more general approach to distill behavior in the event log related to a place in an accepting Petri net and that can be used to discover and insert more precise accepting Petri nets. This approach guarantees preservation of fitness and precision. Additionally, we compare our work to an existing approach on inserting precise models and show that instances exist on which the other approach fails and where our approach succeeds. Finally, we evaluate the ProjectionMiner towards performance and the improvement of precision, generalization and simplicity for event logs from literature and real-life.