Decomposing replay problems : a case study

Conformance checking is an important field in the process mining area. In brief, conformance checking provides us with insights how well a given process model matches a given event log. To gain these insights, we typically try to replay the event log on the process model. However, this replay problem may be complex and, as a result, may take considerable time. To ease the complexity of the replay, we can decompose the given process model into a collection of (smaller) submodels, and associate a (smaller) sublog to every submodel. Instead of replaying the entire event log on the entire process model, we can then replay the corresponding sublog on every submodel, and combine the results. This paper tests this divide-and-conquer approach on a number of process models and event logs while using existing replay techniques. Results show that the decomposed replay may indeed be faster by orders of magnitude, but that success is not guaranteed, as in some cases a smaller model and a smaller log yield a more complex replay problem.