This paper presents a novel method for detecting anomalous vessel behavior in maritime traffic using unsuper-vised clustering techniques like K -Means and Self-Organizing Maps (SOM). Our approach, differing from traditional meth-ods, initially clusters regular trajectories to provide a stronger foundation for anomaly detection. This approach is grounded in the observation that certain trajectories, while not evidently anomalous if compared against a heterogeneous dataset of regular paths, become easily noticeable when compared against clustered regular trajectories. By clustering standard maritime routes, we establish a baseline for normal traffic patterns, iden-tifying anomalies through their deviation from these clusters. Our approach has the potential to be particularly effective in an online setting, as the clustering of regular trajectories is conducted offline. This avoids the necessity of executing clus-tering algorithms online, simplifying the process and enhancing operational efficiency. The paper includes an experimental analysis using real Automatic Identification System (AIS) data and synthetic anomalous vessel trajectories, demonstrating the method's effectiveness.
Anomalous Vessel Behavior Detection via Offline Clustering of Regular Trajectories
Fioravanti C.;Oliva G.;Setola R.
2024-01-01
Abstract
This paper presents a novel method for detecting anomalous vessel behavior in maritime traffic using unsuper-vised clustering techniques like K -Means and Self-Organizing Maps (SOM). Our approach, differing from traditional meth-ods, initially clusters regular trajectories to provide a stronger foundation for anomaly detection. This approach is grounded in the observation that certain trajectories, while not evidently anomalous if compared against a heterogeneous dataset of regular paths, become easily noticeable when compared against clustered regular trajectories. By clustering standard maritime routes, we establish a baseline for normal traffic patterns, iden-tifying anomalies through their deviation from these clusters. Our approach has the potential to be particularly effective in an online setting, as the clustering of regular trajectories is conducted offline. This avoids the necessity of executing clus-tering algorithms online, simplifying the process and enhancing operational efficiency. The paper includes an experimental analysis using real Automatic Identification System (AIS) data and synthetic anomalous vessel trajectories, demonstrating the method's effectiveness.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.