Semi-supervised Log Pattern Detection and Exploration Using Event Concurrence and Contextual Information

Abstract

Process mining offers a variety of techniques for analyzing process execution event logs. Although process discovery algorithms construct end-to-end process models, they often have difficulties dealing with the complexity of real-life event logs. Discovered models may contain either complex or over-generalized fragments, the interpretation of which is difficult, and can result in misleading insights. Detecting and visualizing behavioral patterns instead of creating model structures can reduce complexity and give more accurate insights into recorded behaviors. Unsupervised detection techniques, based on statistical properties of the log only, generate a multitude of patterns and lack domain context. Supervised pattern detection requires a domain expert to specify patterns manually and lacks the event log context. In this paper, we reconcile supervised and unsupervised pattern detection. We visualize the log and help users extract patterns of interest from the log or obtain patterns through unsupervised learning automatically. Pattern matches are visualized in the context of the event log (also showing concurrency and additional contextual information). Earlier patterns can be extended or modified based on the insights. This enables an interactive and iterative approach to identify complex and concrete behavioral patterns in event logs. We implemented our approach in the ProM framework and evaluated the tool using both the BPI Challenge 2012 log of a loan application process and an insurance claims log from a major Australian insurance company.