Multi-perspective Concept Drift Detection: Including the Actor Perspective

Changes in processes manifest as concept drift in event logs. Drift detection aids in analyzing the nature of such change and its impact on the process. Process executions or cases are driven by actors and machines performing the actual work. Actors typically divide and structure their work into tasks—multiple consecutive actions performed together—before handing a case to the next actor. Process changes affect this work division and collaboration, potentially impacting performance and outcomes. However, existing research on concept drift detection from event logs has not yet focused on the behavior of actors. We generalize an existing concept drift detection technique to consider actor behavior and control-flow jointly by using a multi-layered event knowledge graph. We evaluate our proposal by comparing the theoretical properties of the newly defined actor perspective features with existing features and perform an experimental evaluation. The experiments showed actor features to be more robust with on average (up to factor 2.6) stronger signals for concept drift in two real-life datasets. Our approach led to new insights into global process changes, changes in behavior of individual actors, and change in collaborations between actors.