Clustering of maritime trajectories with AIS features for context learning

This paper presents an analysis on Automatic Identification System (AIS) real world ship data to build a system with the capability to extract useful information for an anomaly detection problem. The study focuses on the adjustment of a clustering technique to trajectory data, specifically using a DBSCAN algorithm that is adapted by means of two approaches. On the one hand, the DTW trajectory similarity metric is used to obtain a distance between two trajectories. On the other hand, an extraction of features of interest from each trajectory allowing a summary of the trajectory in a single multidimensional instance. The results show that both approaches are feasible, although not very scalable to larger problems due to the computational complexity of the used algorithms. In addition, the study analyses possible uses of these approaches to existing data mining problems.