We often use an analogy in process mining of desire paths, the tracks made in grass by people walking outside designed lines; unplanned records of the way people actually work. This pretty example is from Sydney University.

But often when we have a dataset, we don’t have the complete dataset, and we can’t see the complete path. We can be missing individual data points, but at other times, we may be missing entire types of data. This is because of the complicated and opportunistic nature of the data we, especially process data, which was often originally recorded for some other purpose. In a process context, this can make analysis of end to end processes difficult or impossible. We call these unobserved subprocesses process voids.

I gave a talk about process voids at the Australian Data Network conference (ADSN 2025), at Sydney Uni this week, based on research done together with Moe Wynn. It shows ways to integrate model information from stakeholders to reason about what data is missing, including highlighting missing subprocesses and an overall coverage metric. It builds on stochastic process mining, skip alignments and skip probabilities. This research is still in progress. If you want to know more about it straight away, you can check out the slides from the talk and the code for the examples.

References

Bär, P., Leemans, S.J.J, Wynn, Moe T. (2025). A Full Picture in Conformance Checking: Efficiently Summarizing All Optimal Alignments. BPM 2025.

Bär, P., Burke, A.T., Leemans, S.J.J, Wynn, Moe T. (2025). Skip Probabilities for Subprocesses. ICPM 2025.

Burke, A., Leemans, S. J. J., & Wynn, M.T. (2021). Discovering Stochastic Process Models By Reduction and Abstraction.

Burke, A., Wynn, M.T. (2025). Process Voids: Data Science Without Data. Talk. Australian Data Science Network 2025.