Predicting the Destination Port of Fishing Vessels

Abstract

Regulating the catch of fishing vessels is crucial for maintaining sustainable fish populations, preventing illegal fishing, and ensuring the quality of the fish being delivered. One effective method of controlling the catch is to have controllers physically present at the port where the catch is being delivered. However, vessels do not always report their destination port in a timely manner, which limits the ability of controllers to regulate the delivered catch. In order to improve the ability of controllers to regulate the catch, this thesis explores how to forecast the destination ports of fishing vessels without relying on their manually transmitted information. We utilize Automatic Identification System (AIS) data to analyze the movement patterns and behaviors of fishing vessels in our dataset, and extend existing work on vessel trajectory predictions using machine learning, to forecast the destination ports of fishing vessels. Additionally, we develop a statistical baseline model to compare our results against. Our results demonstrate that both models correctly predict the destination port of a given vessel in the majority of times, with the accuracy of the machine learning approach increasing as more input data is added. The statistical baseline mode performs better with vessels that do not visit a variety of ports, while the machine learning approach provides a better overall assessment, and thus appears to be the more promising approach. Both models have the potential to be improved considerably by incorporating more input features.