State Matters: Detecting Behavioral Patterns in Object-Centric Process Mining

Enterprise processes involve many interacting objects whose behavior depends on operational states. Object-centric process mining with OCEL 2.0 captures interactions between objects, and state-aware object-centric process mining adds the state evolution of selected objects. Identifying recurring local behavioral patterns that contribute to entering, maintaining, or recovering from undesired states is essential for process analysis and for designing improvement measures. However, detecting these patterns currently relies on manual inspection of state-aware directly-follows graphs, which is complex and does not scale. This paper presents an automated pattern detection approach for state-aware object-centric process mining. Given a leading object type, the method segments its state evolution, represents each segment as an object-centric graph, and aggregates structurally equivalent segments into ranked patterns. The method distinguishes patterns that occur inside a state from patterns that span state changes. In a real-life case study conducted with Europe’s leading pet retailer, the analysis reveals the behavioral patterns most strongly associated with understock and overstock states, providing a finer-grained diagnostic view of process behavior.