Finding Optimal Contextual Parameters for Real-Time Vessel Position Prediction Using Deep Learning

About 80% of world transportation happens at sea. Therefore the safety of vessels, in particularduring vessels’ movement, is crucially important. As different contextual parameters affect vessels’movement, selecting optimal contextual parameters is one of the main changes in vessels’ Context-Aware movement analysis. Toward this end, a Long Short-Term Memory (LSTM) network is usedfor wrapper feature selection to identify optimal contextual parameters for vessels’ movementprediction. To do this, the Automatic Identification System (AIS) dataset from the eastern coast of theUnited States of America collected from December 2017 is used. All possible combinations of threecontextual parameters, including speed, course and vessels’ presence probability in different positionsat sea, were evaluated using the wrapper method in the LSTM network. In all evaluations, 70% ofdata was used for training and the remaining for cross-validation. The results selected speed andpresence probability as optimal contextual parameters for vessel movement prediction. The modeltrained with optimal contextual parameters is 26.98% more accurate than a model trained with allavailable contextual parameters and 16.14% better than a model without contextual parameters.Therefore, selecting optimal parameters from available contextual parameters can help improve theaccuracy of vessels’ predictions. Keywords: Context-Aware, Long Short-Term Memory, AutomaticIdentification System, wrapper, Movement prediction, Context.