Mahalanobis Distance Based DBSCAN For Vessel Near Collision Detection

Mahalanobis Distance-based DBSCAN is a method for detecting potential near-collision among ship using AIS data. In general, the detection of near-collision is very important in preventing ship accidents. DBSCAN plays a critical role in detecting potential collision between maritime ship. Furthermore, its clustering method identifies ship that is in close proximity, showing potential collision risks. The distance between these objects is a fundamental aspect of DBSCAN clustering method. The attributes of latitude, longitude, and heading in AIS data showed correlations and diverse distributions. However, Euclidean distance measurement failed to consider these correlations and variations. To address this limitation, the current study proposed Mahalanobis distance-based DBSCAN clustering to recognize potential near-collision scenarios. The method incorporated receiving AIS data, cleaning the data to remove docked ship, calculating the covariance matrix, computing Mahalanobis distance, performing DBSCAN clustering, and visualizing cluster outcomes. Visualization of cluster, measurement of distance variables, and assessment of silhouette coefficient scores were used to analyze near-collision detection results. Mahalanobis-based DBSCAN effectively generated cluster associated with near-collision scenarios, as shown by the Dr, DCP A, and TCP A values of the ship pairs. Mahalanobis distance showed superior cluster quality compared to Euclidean distance, with scores of 0.291 and 0.125, respectively.