Qualitative Assessment of the Exceeding Behaviour of Discovered Petri Nets

Abstract

In this paper, the precision evaluation of a subclass of Petri nets, called workflow nets (WFN), is addressed. A novel technique for qualitatively assessing the precision is proposed; it estimates the exceeding language of a sound WFN including cycles and silent transitions with respect to the actual language provided by the events log. The estimation, which does not consider the frequency of traces in the log, involves three sub-metrics dealing with a) activity completeness, b) finite behaviour of the model, and c) cyclic behaviour of the model. The compound metric is computed through a partial reachability graph called log-based marking graph (LMG); it is iteratively evaluated by simulating the WFN and adding new behaviour to such a LMG until the computed value converges asymptotically. The proposed technique provides a good approximation to the precision of sound WFNs modelling finite or infinite languages.