An Infrastructure for Cost-Effective Testing of Operational Support Algorithms Based on Colored Petri Nets

摘要

Operational support is a specific type of process mining that assists users while process instances are being executed. Examples are predicting the remaining processing time of a running insurance claim and recommending the action that minimizes the treatment costs of a particular patient. Whereas it is easy to evaluate prediction techniques using cross validation, the evaluation of recommendation techniques is challenging as the recommender influences the execution of the process. It is therefore impossible to simply use historic event data. Therefore, we present an approach where we use a colored Petri net model of user behavior to drive a real workflow system and real implementations of operational support, thereby providing a way of evaluating algorithms for operational support before implementation and a costly test using real users. In this paper, we evaluate algorithms for operational support using different user models. We have implemented our approach using Access/CPN 2.0.`