Predictive modelling of tugboat propulsion loads using machine learning: a case study on Istanbul Strait maritime operation

OpenAlex English

Tugboats are essential to port operations but contribute disproportionately to fuel consumption and emissions, challenging operational efficiency and decarbonization compliance. Accurate engine load prediction is crucial for fuel optimization and regulatory reporting, yet the dynamic nature of tugboat manoeuvres complicates the task. This study evaluates four machine learning algorithms, K-star (K*), M5 Rules, Random Forest (RF), and Multilayer Perceptron (MLP), to predict engine load using real operational data from the Istanbul Strait. Models were validated through 10-fold cross-validation and multiple train-test splits. All algorithms achieved strong predictive performance, with R2 values ranging from 0.94 to 0.9544. K* delivered the highest accuracy and fastest training time, while M5 Rules offered superior interpretability. RF demonstrated competitive accuracy with moderate tuning, and MLP showed consistent behavior despite higher computational effort. These findings provide practical insights for selecting machine learning models in maritime operations, contributing to energy efficiency and emission reduction for harbor craft.