Maritime Craft Navigation Pattern Characterization and Prediction Based on Automated Identification System Data from the Persian Gulf

Maritime planners and analysts seek to characterize vessel behavior and predict future motion in highly congested and contested waterways. This thesis examines historical Automated Identification System (AIS) data from the Persian Gulf over six months. Extensive data preprocessing and outlier detection facilitates the construction of vessel sub-tracks between an origin and destination point. Next, cluster analysis of vessel sub-tracks is used to develop a Bayesian probability of cluster membership for each successive point in the sub-track. This probability, combined with dynamic and static vessel data including ship coordinates, identification number, ship size, and ship type, are used as features in a recurrent neural network model for predicting future ship trajectory. Finally, this thesis identifies potential modeling approaches that may be used to more accurately predict vessel trajectories or identify anomalous vessel behavior near contested waters.