Detecting Anomalies in Healthcare Processes: A K-NN Graph-Based approach

Detecting anomalies in healthcare processes helps identify irregular patterns that may point to medical errors, inefficiencies, or departures from clinical protocols. As event logs become more common in clinical systems, they offer a valuable resource for understanding and improving how healthcare is delivered. This increases the relevance of process mining, which supports the analysis and improvement of real-world workflows by examining the sequence of events recorded in these logs. However, working with real event data can be difficult due to its complexity and variability. To manage this, we use graph-based models that highlight the relationships between activities without generating overly complicated, hard-to-read process maps. In this study, we focus on K-Nearest Neighbors (K-NN) graph modeling, which does not require setting the number of clusters in advance. This flexibility allows for a more detailed view of the process landscape and helps identify subtle variations or unusual behaviors. Using a hospital billing dataset, we show that graph-based methods—especially K-NN—can better group typical and unusual patterns when compared to more traditional techniques. These results support the use of graph models as a practical tool for reviewing and improving healthcare processes, with the potential to support better management and patient outcomes.