Enhancing Behavioural Anomaly Detection Under Concept Drift within Healthcare Sector: Application of Change Point Detection and Batch Learning

The dynamic nature of human behaviour poses challenges for behavioural anomaly detection models that can be impacted by concept drift. This experimental study employs the Aruba real-world dataset obtained from CASAS to examine the effectiveness of using Change Point Detection and Batch Learning in adapting Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and Autoencoder models. Results demonstrate that the proposed approach surpasses the baseline of no adaptation, yielding an average improvement of 10.46% for DBSCAN, with a performance of 3.96% points higher than the benchmark regular adaptation. Similarly, Autoencoder achieves an average improvement of 4.01%, with a performance 4.11% points higher than the benchmark. The findings suggest the need for increased attention to address concept drift in behavioural anomaly detection and highlight the potential benefits of enhancing detection capabilities in the presence of concept drift.