Finding Structure in the Unstructured: Hybrid Feature Set Clustering for Process Discovery

Process discovery is widely used in business process intelligence to reconstruct process models from event logs recorded by information systems. With the increase of complexity and flexibility of processes, it is getting more and more challenging for discovery algorithms to generate accurate and comprehensive models. Trace clustering aims to overcome this issue by splitting event logs into smaller behavioral similar sub-logs. From these sub-logs more accurate and comprehensive process models can be reconstructed. In this paper, we propose a novel clustering approach that uses frequent itemset mining on the case attributes to also reveal relationships on the data perspective. Our approach includes this additional knowledge as well as optimizes the fitness of the underlying process models of each cluster to generate accurate clustering results. We compare our method with six other clustering methods and evaluate our approach using synthetic and real-life event logs