Estimating weather margin seasonality in shipping using machine learning

Abstract

Accurate predictions of fuel consumption are an essential tool in the pricing of forward cargo contracts. This thesis develops a predictive model for fuel consumption using noon report data from Handysize and Supramax vessels. In the process, we employ a wide selection of machine learning algorithms, including decision trees, shrinkage models, and an artificial neural network. Furthermore, we replace all weather and oceanographic variables with third-party data. The replacement ensures the model is independent of noon report weather data and allows us to generate predictions using historical weather conditions from the last decades. The trained models are used to study the seasonal patterns of weather margins for two case routes. Estimated weather margins and fuel consumption may be used by chartering managers to improve cost predictions and facilitate more profitable contract selection.