On Making Informed Decisions Regarding Clustering Event Logs in Process Discovery

Process discovery is intended to machine‐learn process models by analysing event logs that record how real processes work. The resulting models tend to be complex because they must account for all the variability in the event logs. It is agreed that clustering the input event log prior to discovery may result in better models. Unfortunately, using a clustering approach has a negative computational impact on the time and the resources required to preprocess the event logs. Furthermore, the literature does not provide a fully developed study regarding their impact on the quality of the resulting process models. This makes it difficult for an expert to make informed decisions regarding how a particular event log must be clustered to reduce preprocessing effort and increase model quality. In this paper, we explore how to address the previous problem by means of a recommender that builds on a sound statistical procedure that is complemented with a visual analysis when necessary. Our proposal is intended to guide experts' work more efficiently and effectively by making informed decisions regarding how to cluster their event logs in process discovery.