Ship Sub-Trajectories Clustering: A Comparative Study on DBSCAN and Spectral Clustering with Dimensionality Reduction

Maritime transportation, handling over 80% of global trade, is critical to the world economy. Automatic Identification System (AIS) data provides extensive static and dynamic information of vessels, enabling trajectory reconstruction and vessel behavior analysis. Recently, trajectory clustering has become a key method for analyzing maritime traffic, offering valuable insights to improve traffic management and operational efficiency. This research aims to investigate how to effectively cluster ship sub-trajectories derived from AIS data by comparing two machine learning clustering algorithms, Density-based spatial clustering of applications with noise (DBSCAN) and spectral clustering, with a focus on improving data quality, extracting key dynamic features, and evaluating the effect of dimensionality reduction on clustering performance. Clustering sub-trajectories can help reveal localized navigation patterns and movement behaviors. The study implemented the proposed methods for tankers and cargo ships (with AIS data from 2022) in a Y-shaped channel in the Sabine-Neches Waterway (SNWW) in Southeast Texas, where the busiest docks are located. Finally, clustering performance was evaluated with the silhouette coefficient (SC), Davies–Bouldin Index (DBI), and Joint Performance Index (JPI), respectively. Experimental results show that DBSCAN effectively identifies dense, overlapping trajectory clusters and labels noise, while the spectral clustering algorithm detects subtle behavioral differences but struggles with less cohesive clusters, and does not explicitly handle noise.