Ship Movement Prediction Using k-NN Method

Abstract

Trajectories of ships travelling in the Gulf of Finland were predicted using the k-Nearest Neighbours (k-NNs) method. Automatic Identification System (AIS) data gathered via open interface of the Finnish Transport Agency were used. The results will be exploited in a route optimization task for an emission control boat. The task requires prediction several hours ahead with reasonable accuracy. The idea is to compare the trajectories of a new ship and historical ships within a comparison area. The future behaviour of the new ship was estimated with the k-nearest neighbours. The performance of the method as well as the hyper parameters (nearest neighbours, k, and a weighting parameter α) of the proposed model were optimized using nested leave-one-out crossvalidation approach. The method enables the prediction within minutes' accuracy in time and less than 2 km in location several hours ahead, which is more than satisfactory for the route optimization purposes.