A novel machine learning approach to analyzing geospatial vessel patterns using AIS data

Abstract

In the maritime environment, the Automatic Identification System (AIS) contains information related to vessel trajectories that can be used to detect unusual maritime occurrences and maritime traffic patterns. To detect such occurrences with supervised learning methods the AIS messages must be manually annotated, which can be a demanding process. Therefore, unsupervised methods are used to identify anomalous traffic patterns based on vessel trajectories. Typically, dense regions of maritime activity are studied to capture common traffic patterns which help identify trajectories that do not follow the norm. However, these approaches cannot detect anomalous behaviors along common pathways or incorporate time-related events into the analysis. Such challenges motivate the approach taken in this work by using auto-regressive techniques to model vessel trajectories and clustering analyses to explore behavior patterns of vessels. Results confirm that the Auto-regressive Integrated Moving Average (ARIMA) and Ornstein-Uhlenbeck (OU) processes are able to model the trajectories and can be used with density-based spatial clustering of applications with noise (DBSCAN), hierarchical clustering (HC), and spectral clustering (SC) to identify different behavioral patterns.