A Preliminary Approach to Probabilistic Process Discovery

Process discovery is a fundamental task in Process Mining, concerned with the extraction of process models from event logs. Alongside traditional deterministic approaches, recent works have started to consider probabilistic declarative process models, where model constraints are learnt and then annotated with weights to account for uncertainty and variability in observed behavior, as well as noise. In this setting, weights have been interpreted, up to now, in terms of the constraints’ support in the observed traces. However, the probabilistic process discovery task remains quite unexplored, often relying on the extraction of constraints by deterministic miners. In this paper, we report preliminary results on the use of PASCAL, a Probabilistic Inductive Constraint Logic algorithm, based on the Distribution Semantics, which is able to discover probabilistic declarative process models, i.e., integrity constraints annotated with a probability, directly from process traces labelled as positive and negative, without the need of a two-step approach. Preliminary results on two synthetic logs indicate that the learned theories recover most of the constraints underlying the process models.