The focus in the field of process mining, andprocess discovery in particular, has thus far been onexploring and describing event data by the means ofmodels. Since the obtained models are often directly basedon a sample of event data, the question whether they alsoapply to the real process typically remains unanswered. Asthe underlying process is unknown in real life, there is aneed for unbiased estimators to assess the system-quality ofa discovered model, and subsequently make assertionsabout the process. In this paper, an experiment is describedand discussed to analyze whether existing fitness, precisionand generalization metrics can be used as unbiased esti-mators of system fitness and system precision. The resultsshow that important biases exist, which makes it currentlynearly impossible to objectively measure the ability of amodel to represent the system.
Depaire, Benoıˆt and Janssenswillen, Gert
"Towards Confirmatory Process Discovery: Making AssertionsAbout the Underlying System,"
Business & Information Systems Engineering:
Vol. 61: Iss. 6, 713-728.
Available at: https://aisel.aisnet.org/bise/vol61/iss6/6