An Enhanced Ship-Speed Prediction Model With Stacking Ensemble Learning

In the field of maritime transportation, precise accurate prediction of ship-speed is paramount for route planning, ship scheduling, and navigational safety. However, the difficulty of the prediction lies in the fact that ship-speed is influenced by numerous complex and variable factors, and traditional speed prediction methods often struggle to attain the desired accuracy in the presence of complex maritime environments and ship characteristics. To address this challenge, this paper aims to develop a robust ship-speed prediction model capable of effectively capturing complex data relationships and improving prediction accuracy and stability. A stacking ensemble learning model is proposed, integrating extra trees (ET), random forest (RF), categorical boosting (CatBoost), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and histogram gradient boosting (HGBT) as base learners, with support vector machine (SVM) as the meta-learner to leverage diverse models’ strengths. A standardized workflow for data cleaning, multi-source data fusion, and feature engineering is established. Additionally, the SHapley Additive exPlanation (SHAP) is introduced for model interpretability. Experiments with historical trajectory data from five ships and meteorological-oceanographic data show that the stacking model outperforms single models in prediction accuracy and stability. SHAP analysis reveals that ship course and wave height are key influencing factors, with their impact varying across different navigation scenarios. The proposed model enhances operational efficiency, safety, and decision-making in the maritime industry by providing reliable speed predictions and interpretable insights.