Mining Expressive Performance Models out of Low-level Multidimensional Process Logs

Process Mining techniques have been gaining attention, owing to their potentiality to extract compact process models from massive logs. Traditionally focused on workflows, these techniques tend to rely on a clear specification of process tasks, assumed to be referred explicitly by the logs. This limits however their applicability to many real-life BPM environments (e.g. issue tracking systems) where the traced events do not match any task, but yet keep lots of context data. To make the application of (predictive) process mining to such logs more effective and easier, a novel approach is devised, where the discovery of different execution scenarios is combined with the automatic abstraction of log events. The approach was integrated in a BPA system, also supporting the evaluation of discovered models and OLAP-like analyses. Tested on real-life data, the approach achieved compelling prediction accuracy w.r.t. state-of-theart methods, and discovered interesting descriptions for both activities and process variants.