Stochastic process mining: Earth movers' stochastic conformance. Issue 102 (December 2021)

• Record Type: Journal Article
• Authors: Leemans, Sander J.J.; van der Aalst, Wil M.P.; Brockhoff, Tobias; Polyvyanyy, Artem
• Is Part Of: Information systems. Issue 102(2021)
• Publication Date: 2021-12
• Subjects: Process mining -- Conformance checking -- Stochastic process mining; Database management -- Periodicals; Electronic data processing -- Periodicals; Bases de données -- Gestion -- Périodiques; Informatique -- Périodiques; Database management; Electronic data processing; Periodicals; 005.7
• DOI: 10.1016/j.is.2021.101724
• Languages: English
• ISSNs: 0306-4379

Abstract

Abstract: Initially, process mining focused on discovering process models from event data, but in recent years the use and importance of conformance checking has increased. Conformance checking aims to uncover differences between a process model and an event log. Many conformance checking techniques and measures have been proposed. Typically, these take into account the frequencies of traces in the event log, but do not consider the probabilities of these traces in the model. This asymmetry leads to various complications. Therefore, we define conformance for stochastic process models taking into account frequencies and routing probabilities. We use the earth movers' distance between stochastic languages representing models and logs as an intuitive conformance notion. In this paper, we show that this form of stochastic conformance checking enables detailed diagnostics projected on both model and log. To pinpoint differences and relate these to specific model elements, we extend the so-called 'reallocation matrix' to consider paths. The approach has been implemented in ProM and our evaluations show that stochastic conformance checking is possible in real-life settings. Highlights: Frequencies – how often paths are used in a process – matter in process mining. This stochastic perspective is often ignored in conformance checking techniques. We introduce stochastic log–log, log–model and model–model comparison techniques. These are feasible, and enable detailed insights into processes' differences.