Complexity analysis of process models considering frequency of traveled paths

fordanic

In ProM 5.2 there is a plugin available for running Petri Net Complexity Analysis, listing metrics such as: workflow, number of AND/XOR-joins/splits and number of arcs/places/transitions. As such, this plugin only looks at the process model represented as a Petri Net, but pays no attention to the behavior as described by for instance the corresponding event log.

This is a problem to me in the case of an event log where the majority of the traces corresponds to a very simple process model but then there are a few traces which correspond to a very complex process model. Many process discovery algorithms will see both process models and include them in the final process model. Hence, when running a complexity analysis on this the process model the obtained metrics will correspond to a complex process model rather than the simple process as described by the majority of the traces in the event log.

Any suggestions for how to deal with this? Maybe in terms of suggestions for how a complexity module could be extended to consider frequency of traveled paths in the process model.

JBuijs

Dear Fordanic,

(sorry for the late reply, hope it is still relevant)

The ETM (Evoltutionary Tree Miner) algorithm is capable of incorporating several quality aspects in the discovery phase. Furthermore, the quality aspects/metrics are easily extensible or added, so you could write your own and let thet ETM find a simple process model, sacrificing replay fitness.

My suggestion therefore is that process model complexity/simplicity cannot be seen in isolation of replay fitness, precision and generalizaton.