DB-Drift: Concept drift aware density-based anomaly detection for maritime trajectories
Amelia A. Henriksen · 2023
Anomalies in maritime surveillance operations are often high-risk, and need to be detected quickly from real-world, incoming data sources. Hence it is critical that we develop unsupervised anomaly detection algorithms that both operate on a data stream and adapt to it. Real-world maritime data streams involve multiple, intersecting forms of concept drift, meaning that the underlying data distribution changes over time. We introduce DB-Drift, a novel algorithm for adapting existing density-based unsupervised anomaly detection pipelines to handle gradual and seasonal drift simultaneously.