Metaheuristic-driven optimization of machine learning models for predicting principal dimensions of container ships

Abstract Accurate prediction of the principal dimensions of container ships is essential for preliminary design and operational planning in the maritime industry. This study explores the integration of metaheuristic optimization algorithms with machine learning models to enhance the predictive performance for four key target variables: length (L), beam (B), draft (T), and block coefficient (CB). Three baseline models are extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and support vector regression (SVR) were individually optimized using grey wolf optimizer (GWO), whale optimization algorithm (WOA), and particle swarm optimization (PSO) to form hybrid prediction frameworks. Performance evaluation was conducted using R 2 , RMSE, and MAE metrics across training, validation, and testing datasets. The results demonstrate that the hybrid models consistently outperform their non-optimized counterparts, achieving higher accuracy and better generalization. Notably, GWO yielded stable improvements across all models and targets, while WOA and PSO showed target-specific enhancements, particularly in Beam and Draft predictions, respectively. This work highlights the critical role of algorithm selection in model optimization and confirms the potential of metaheuristic-augmented machine learning models in the context of ship design automation and early stage parameter estimation.