Learning Abnormal Vessel Behaviour from AIS Data with Bayesian Networks at Two Time Scales
Steven Mascaro, Kevin B. Korb, Ann E. Nicholson · 2010
In recent years, electronic vessel tracking has provided abundant data on vessel movements to surveillance authorities. Researchers have begun looking at the use of this data for anomaly detection using a wide variety of data mining techniques. Here we tackle anomaly detection with Bayesian Networks, training them with real world AIS data and producing models at two different time scales — both moment to moment and for the track as a whole. The networks also incorporate additional real world data, including weather, vessel details and details about vessel interactions. We find that the generated networks are quite easy to examine and verify despite incorporating a large number of variables; that combining models at the two different scales improves performance in a variety of cases; and, ultimately, that Bayesian Networks prove a promising approach to anomaly assessment and detection. 1