Beyond Flattening: Detecting Concurrency Anomalies Using K-NN Graph-Based Modeling in Object-Centric Event Logs

Detecting anomalous executions is essential in today’s dynamic and diverse business environments. It plays a pivotal role in identifying inefficiencies, ensuring compliance, and mitigating risks associated with deviations from standard procedures. Traditional process mining techniques generally assume a linear sequence of events. However, real-world processes often present concurrency, characterized by the parallel execution of multiple activities or cases and complex interactions among events. These behaviors are not mapped by conventional linear models, this way, not accurately capturing the dynamic nature of process flows. To tackle this challenge, this study proposes a new approach for detecting concurrency anomalies using a K-NN graph-based model, overcoming the traditional flattening method. In our experiments, we explored object-centric event logs with different types of concurrency anomalies and compared them to the traditional flattening procedure. Our proposal was able to provide comprehensive and precise communities (clusters) of anomalous variants compared to the baseline.