Comparisons of Machine Learning Methods in Ship Speed Prediction Based on Shipboard Observation

This study presents a novel approach to predicting ship speed based on real-time voyage observation data, aiming to enhance maritime safety and operational efficiency. Observational data from a 20,000-ton bulk carrier, including variables such as latitude, longitude, GPS orientation, wind direction, wind speed, and main engine parameters, were collected and preprocessed to mitigate noise and handle missing values. Six machine learning models—the Backpropagation (BP) Neural Network, Support Vector Regression (SVR), Decision Tree (DT), Random Forest (RF), XGBoost, and LightGBM—were employed to develop predictive models. Among these, the LightGBM model demonstrated the highest prediction accuracy, achieving a Root Mean Squared Error (RMSE) of 0.188, Mean Absolute Error (MAE) of 0.149, and a coefficient of determination (R2) of 0.978. The results highlight the potential of the LightGBM model in optimizing ship navigation and improving maritime operational efficiency. These findings offer a reliable foundation for further advancements in predictive maritime technologies and route optimization.