The visualization of models is essential for user-friendly human-machine interactions during Process Mining. A simple graphical representation contributes to give intuitive information about the behavior of a system. However, complex systems cannot always be represented with succinct models that can be easily visualized. Quality-preserving model simplifications can be of paramount importance to alleviate the complexity of finding useful and attractive visualizations. This paper presents a collection of log-based techniques to simplify process models. The techniques trade off visual-friendly properties with quality metrics related to logs, such as fitness and precision, to avoid degrading the resulting model. The algorithms, either cast as optimization problems or heuristically guided, find simplified versions of the initial process model, and can be applied in the final stage of the process mining life-cycle, between the discovery of a process model and the deployment to the final user. A tool has been developed and tested on large logs, producing simplified process models that are one order of magnitude smaller while keeping fitness and precision under reasonable margins.

Peer Reviewed