Identifying Process Graphs Properties with Network Science Metrics

Process Mining graphs are models that represent business processes. These graphs have been used in some contexts as social networks and process concept drift. However, they have scarcely been studied in the context of network science as graphs with particular properties. In this work, we used network science metrics and machine learning models to distinguish process graphs from diverse non-process graphs belonging to social and random models. We performed our experiments with a real dataset containing multiple process logs from a Brazilian justice system. We generated non-process graphs with Barabási, Duplication-Divergence, Erdõs-Rényi, Gaussian Random Partition, and Newman Watts Strogatz generators. Our results suggest that the metrics used are highly efficient to distinguish among the analysed graphs. The process graphs presented particular characteristics such as higher clustering coefficient and lower assortativity than non-process graphs. These findings may encourage the usage of network science metrics and machine learning models for process mining challenges in big data logs.