Earth Movers' Stochastic Conformance Checking

摘要

Process Mining aims to support Business Process Management (BPM) by extracting information about processes from real-life process executions recorded in event logs. In particular, conformance checking aims to measure the quality of a process model by quantifying differences between the model and an event log or another model. Even though event logs provide insights into the likelihood of observed behaviour, most state-of-the-art conformance checking techniques ignore this point of view. In this paper, we propose a conformance measure that considers the stochastic characteristics of both the event log and the process model. It is based on the "earth movers' distance" and measures the effort to transform the distributions of traces of the event log into the distribution of traces of the model. We formalize this intuitive conformance metric and provide an approximation and a simplified variant. The latter two have been implemented in ProM and we evaluate them using several real-life examples.