Finding process variants in event logs: (Short Paper)

Abstract

The analysis of event data is particularly challenging when there is a lot of variability. Existing approaches can detect variants in very specific settings (e.g., changes of control-flow over time), or do not use statistical testing to decide whether a variant is relevant or not. In this paper, we introduce an unsupervised and generic technique to detect significant variants in event logs by applying existing, well-proven data mining techniques for recursive partitioning driven by conditional inference over event attributes. The approach has been fully implemented and is freely available as a ProM plugin. Finally, we validated our approach by applying it to a real-life event log obtained from a multinational Spanish telecommunications and broadband company, obtaining valuable insights directly from the event data.