A comparison of advanced regression techniques for predicting ship CO2 emissions

The new EU Regulation urges shipping operators to set up systems for the monitoring, reporting, and verification of CO2 emissions. Indeed, new monitoring data acquisition systems installed on modern ships have brought a navigation data overload that needs to be correctly handled to make proper decisions about their operation. However, in today's market, there is no standard solution or method available that can be robustly adopted in real environments for the shipping industry. In view of the novel attempts for solving this issue proposed by statisticians, marine engineers, and practitioners, this paper presents an extensive comparison of several regression techniques that can exploit the navigation information usually available in modern ships: variable selection methods, penalized regression methods, latent variable methods and tree-based ensemble methods. The comparison is made by means of operational data collected on a Ro-Pax cruise ship owned by the Italian shipping company Grimaldi Group. The goal of this analysis is twofold: (1) to identify methodologies with more potential at analyzing the data collected from this shipping industry scenario and (2) to develop a predictive model for CO2 emissions with good characteristics of accuracy and robustness.