Forecasting Dry Bulk Vessel Flows using AIS Data and Machine Learning

Abstract

This thesis explores the application of machine learning techniques for forecasting dry bulk vessel flows using Automatic Identification System (AIS) data. The study focuses on developing effective preprocessing methods to extract salient features from AIS data and evaluating the performance of state-of-the-art machine learning models, specifically SegRNN and PatchTST, in predicting large-scale shipping trends. A comprehensive preprocessing pipeline is developed to clean, filter, and aggregate the raw AIS data into a structured format suitable for machine learning tasks. The pipeline involves data cleaning, feature engineering, and spatial aggregation using the H3 library to discretise the continuous space into hexagonal cells. The preprocessed data is then used to train and evaluate three machine learning models: nLinear, SegRNN, and PatchTST. The comparative analysis of the models reveals interesting insights into their performance across different vessel types. While the performance of the models leaves room for improvement, the study provides a solid foundation for further research and development. This thesis contributes to the field of shipping trend forecasting by presenting a comprehensive preprocessing approach for AIS data and evaluating state-of-the-art machine learning models. The findings and insights obtained can guide future research.