Behavioral learning of vessel types with fuzzy-rough decision trees

A reliable and efficient characterization of vessel activities along coastal regions is of crucial importance for maritime domain awareness. With increased navigational flows across all waterways and the worldwide dissemination of active and passive vessel tracking modalities, learning a vessel's behavior is becoming a strategic priority for maritime operators and decision makers. In this paper, we propose an interpretable computational model based on fuzzy-rough decision trees (FRDTs) to predict the vessel type given a summary vector in the form of descriptive track features that include kinematic, static and environmental information. The track summaries are generated from the fusion of Automatic Identification System (AIS), Synthetic Aperture Radar (SAR) and Canada weather reports. Our methodology uses fuzzy rough sets to discard irrelevant features on the basis of their dependency of the vessel type, prior to the iterative construction of the FRDT. Empirical results with a real-world data set in the east coast of North America confirm that the proposed approach is able to accurately assign the correct label (i.e., type) to previously unseen vessels in over 80% of the cases.