Identifying Cohorts: Recommending Drill-Downs Based on Differences in Behaviour for Process Mining

Process mining aims to obtain insights from event logs to improve business processes. In complex environments with large variances in process behaviour, analysing and making sense of such complex processes becomes challenging. Insights in such processes can be obtained by identifying sub-groups of traces (cohorts) and studying their differences. In this paper, we introduce a new framework that elicits features from trace attributes, measures the stochastic distance between cohorts defined by sets of these features, and presents this landscape of sets of features and their influence on process behaviour to users. Our framework differs from existing work in that it can take many aspects of behaviour into account, including the ordering of activities in traces (control flow), the relative frequency of traces (stochastic perspective), and cost. The framework has been instantiated and implemented, has been evaluated for feasibility on multiple publicly available real-life event logs, and evaluated on real-life case studies in two Australian universities.