Abstract:

Process modeling is an effective methodology for visually representing workflows in enterprises. However, process outcomes lack quality if relevant details are missing in process models. We have experienced that such details are often not initially known at all or are omitted in the modeling phase to reduce model complexity. Modelers often lack knowledge regarding the relevance of specific details concerning process success leading to an inadvertent exclusion of essential process information. For this reason, methods are required to analyze existing process
models and to identify initially hidden but relevant process details (RPDs). Previous approaches extract RPDs from image data recorded during process execution. These techniques assume demanding prerequisites like the availability of large amounts of execution data, which may not be feasible in small enterprises. To address this challenge, we propose an alternative approach using Association Rule Mining to extract relevant process information from images. We extensively evaluate different algorithms in two use cases inspired by a real manufacturing process. Our experiments confirm the applicability of our novel approach and provide recommendations on the most suitable algorithms for mining RPDs.