PROJECT TITLE :

Measuring Fitness and Precision of Automatically Discovered Process Models: A Principled and Scalable Approach

ABSTRACT:

We are able to generate a process model by using automated process discovery techniques, and the model can be derived from an event log that is a collection of business process execution traces. The quality of the process models that are generated by these techniques can be evaluated with respect to a number of criteria, including fitness, which captures the degree to which the generated process model is able to recognize the traces in the event log, and precision, which captures the degree to which the behavior that is allowed by the process model is observed in the event log. Both of these criteria can be used to evaluate the quality of the process models. In the research that has been done, a wide variety of precision and fitness measures have been suggested. Existing measures in this field, however, do not fulfill basic monotonicity properties and/or they suffer from scalability issues when applied to models discovered from real-life event logs. This is a problem because monotonicity is a fundamental property of this field. This article presents a family of fitness and precision measures that are derived from the concept of contrasting the kth order Markovian abstraction of a process model with that of an event log. These measures are based on the idea that a more accurate model will have a higher degree of accuracy. This family of measurements is shown to satisfy the aforementioned properties when appropriate values of k are selected, as the article demonstrates. An empirical evaluation demonstrates that representative examples of this family of measures produce results that are intuitive when applied to a synthetic dataset of model-log pairs, while at the same time outperforming existing measures of fitness and precision in terms of execution times on real-world event logs.