Explainable machine learning-based prediction of fuel consumption in ship main engines using operational data

Abstract

A significant percentage of fuel consumption and emissions from transportation activities is related to maritime transportation. Hence, accurate prediction models for fuel consumption are quite important. Machine learning offers a data-driven approach to improving fuel consumption prediction, thereby promoting environmental sustainability, lowering operational costs, and enhancing financial viability. This work explores several machine learning approaches by employing statistical measures, including mean squared error (MSE), coefficient of determination (R²), and Kling-Gupta efficiency (KGE), to develop main engine fuel consumption (MEFC) prediction models. Hyperparameter optimization via grid search was conducted to improve the generalizability and robustness of the models. With the lowest test MSE (0.69), a robust testing R² (0.9867), and a high KGE (0.9681), the Random Forests proved to be the most appropriate model for MEFC modeling among all others. Extreme Gradient Boosting followed closely with competitive accuracy, with MSE values of 0.75 and a robust testing R² (0.9856). Using Shapley additive explanations and Local interpretable model-agnostic explanations, this study improves model interpretability even more and indicates that main engine speed and wind speed were revealed to be the most important factors controlling MEFC. Explainable artificial intelligence techniques offer transparency in decision-making, thereby helping marine operators maximize fuel economy. Employing reliable and interpretable predictive modeling, this study offers insightful information for sustainable shipping, hence lowering operating costs and emissions.