bupaRminer: robust discovery of BPMN-models in the bupaR ecosystem

While many advances have been made in the field of process discovery, the use of process discovery algorithms in commercial applications of process mining remains low. State-of-the-art process discovery algorithms still suffer from one or more important disadvantages, such as no soundness guarantee, sensitivity to noise, overly complex models, or inadequate balance between fitness and precision.

In this talk, we propose a new process discovery algorithm - bupaRminer (BM) - that operates in a two-staged approach. In the first stage, the algorithm estimates likely relationships between pairs of activities, the result of which can be seen as a declarative model. In the second stage, these relationships are aggregated into a sound BPMN model. Analysis shows that while Splitminer and Inductive Miner outperform BM in terms of either precision or fitness, respectively, BM is able to achieve a higher F-score on average. Furthermore, BM is found to be more resilient to noisy logs.