Since the introduction of the Automatic Identification System (AIS), AIS data has proven to be a valuable source of ship behavior analysis using data mining. It records ship position, speed and other behavior attributes at specific time intervals in all voyages at sea and in ports. However, the current studies in ship behavior analyze the behavior patterns either with a subjective choice of classification for behavior differences among the groups of ships or without any classification at all. In order to fill this gap, a new methodology for ship classification in ports based on behavior clustering is developed by analyzing AIS data from the port of Rotterdam. Besides a proper data preparation, the proposed methodology consists of two steps: step I, clustering ship behavior in a port area and identifying the characteristics of the clusters; step II, classifying ships to such behavior clusters based on the ship characteristics. The clustering results present both the behavior patterns and the behavior change patterns for ship path and speed over ground, which are the dominant behavior attributes for ships in ports. Some patterns of integral ship behavior can also be revealed by investigating the correlation between the two behavior attributes. Our research has shown that length and beam can be adopted as explanatory variable to classify ships to the corresponding behavior clusters. The classifiers are developed based on both unsupervised discretization (equal width binning) and supervised discretization (Chi2). The performances of classifiers are compared by three evaluation metrics, including Average Accuracy, F 1 score, and AUC. We found that the classification based on multi-criteria is more accurate than using a single criterion. The classifications based on Chi2 discretization outperform the ones with equal width discretization. The outcome leads to a systematic understanding of ship behavior in a port area and can be used to predict the ship behavior pattern based on their characteristics and simulate the ship behavior.
Original languageEnglish
Pages (from-to)176-187
Number of pages12
JournalOcean Engineering
Volume175
DOIs
Publication statusPublished - 2019

    Research areas

  • AIS data, Behavior clustering, Data mining, Ports and waterways, Ship classification

ID: 51646495