Destination Prediction of Oil Tankers Using Graph Abstractions and Recurrent Neural Networks

Predicting the destination of vessels in the maritime industry
is a problem that has seen sustained research over the last few years
fuelled by an increase in the availability of Automatic Identification System
(AIS) data. The problem is inherently difficult due to the nature of
the maritime domain. In this paper, we focus on a subset of the maritime
industry - the oil transportation business - which complicates the problem
of destination prediction further, as the oil transportation market
is highly dynamic. We propose a novel model, inspired by research on
destination prediction and anomaly detection, for predicting the destination
port- and region of oil tankers. In particular, our approach utilises a
graph abstraction for aggregation of global oil tanker traffic and feature
engineering, and Recurrent Neural Network models for the final port- or
region destination prediction. Our experiments show promising results
with the final model obtaining an accuracy score of 41% and 87.1% on a
destination port- and region basis respectively. While some related works
obtain higher accuracy results - notably 97% port destination prediction
accuracy - the results are not directly comparable, as no related literature
found deals with the problem of predicting oil tanker destination on
a global scale specifically.