Using Multi-Level Information in Hierarchical Process Mining: Balancing Behavioural Quality and Model Complexity

Process mining techniques aim to derive knowledge of the execution of processes, by means of automated analysis of behaviour recorded in event logs. A well-known challenge in process mining is to strike an adequate balance between the behavioural quality of a discovered model compared to the event log and the model's complexity as perceived by stakeholders. At the same time, events typically contain multiple attributes related to parts of the process at different levels of abstraction, which are often ignored by existing process mining techniques, resulting in either highly complex and/or incomprehensible process mining results. This paper addresses this problem by extending process mining to use event-level attributes readily available in event logs. We introduce (1) the concept of multi-level logs and generalise existing hierarchical process models, which support multiple modelling formalisms and notions of activities in a single model, (2) a framework, instantiation and implementation for process discovery of hierarchical models, and (3) a corresponding conformance checking technique. The resulting framework has been implemented as a plug-in of the open-source process mining framework ProM, and has been evaluated qualitatively and quantitatively using multiple real-life event logs.