Evolutionary Computation Based Discovery of Hierarchical Business Process Models

Abstract

Business process models that describe how the execution of work in a business is structured are an important asset of modern enterprises. They serve as documentation, and, if easily understandable, allow process stakeholders to make better decisions on the business process. Traditionally, these models have been created manually after analyzing the process, which can lead to outdated information when changes are introduced into the process. Today, information systems connected to the business processes log event data reflecting the real execution of the processes, and process discovery techniques have been developed to automatically extract models from these event logs. Most of these techniques discover well formalized models such as Petri nets, which can be hard to understand in case of larger process models. The evolutionary computation based approach presented in this paper discovers process models complying to the specification of BPMN, one of the most used but not well formalized notations for documenting business processes. Our approach limits the set of possible process models to hierarchically structured models, and therefore facilitates well structured and simple results. An evaluation with eight event logs shows that, despite the limitation to well structured and simple models, the approach delivers competitive results when compared with other process discovery techniques.