Localized Conformance and Performance Analysis based on Event Data: Diagnosing Individual Places
BSc Thesis Project
Title: Localized Conformance and Performance Analysis based on Event Data: Diagnosing Individual Places
Author: Daniel Tacke genannt Unterberg
1st Examiner: Prof.dr.ir. Wil M.P. van der Aalst
2nd Examiner: Prof. Dr. rer. nat. Martin Grohe
Answering questions about the processes behind logged event data is what process mining is all about. To fully uncover complex patterns and problems in the process, several perspectives on the data like conformance and performance, need to be combined. The local process context (termed as busyness in the thesis), like how many cases are currently active in a certain state, is also useful to consider. Furthermore, since some problems may only occur at certain times or stages in the process, the analysis needs to be fine grained. Most state-of-the-art approaches average over the entire process duration which obscures many weaker rare patterns.
To consider the different perspectives in higher detail, we propose an approach to localize conformance and performance to individual places in a petrinet model and discretized time intervals over the process duration. It is based on token sojourn times and move on model counting on aligned traces. We define metrics for fitness, performance and busyness in this localized setting and provide an interactive implementation as a ProM plugin. On top of a small performance analysis, the approach is validated on a synthetic and real-life dataset.