Machine learning as a decision support system in capesize route optimization : predicting optimal route selection using recurrent neural networks and extreme gradient boosting
Abstract
In this thesis, we investigate the applicability of machine learning to predict optimal route
decisions for Capesize vessels. Our approach uses windows of historical macroeconomic
and market-specific variables to form a time-series classification problem. We fit a
Long-Short-Term-Memory neural network and an Extreme Gradient Boosting model on
historical optimal trip choices and predict an out-of-sample routing strategy. By relying
on historical input and no knowledge of the future, we can compute possible economic
gains, and evaluate the viability of machine learning as a decision support system in route
optimization.
In a simplified scenario, with two geographically different roundtrips, we evaluate
cumulative earnings throughout a three-year period. Our findings suggest the machine
learning methods can outperform an approximation of the average earnings of market
participants by almost 11%. Our thesis contributes to the existing literature on spatial
efficiency and routing optimization in the dry bulk market and provides insights into a
possible method of using machine learning in out-of-sample route prediction.