Abstract: Addressing the complexities and the lack of real-time capability in calculating the metacentric height（G_M）of ships,a method for real-time prediction of a ship’s G_M based on an improved Radial Basis Function Neural Network（RBFNN）is proposed. Firstly,leave-one-out cross-validation and early stopping strategies are introduced to optimize the RBF,enhancing the generalization performance of the model. Secondly,the passenger-cargo Ro-ro ship "Zijing 11" navigating in Qiongzhou Strait is selected as the research subject,and four factors closely related to G_M are chosen as input features for the neural network through grey relational analysis. Finally,the G_M calculated using an empirical formula is taken as the expected value and compared with the predicted values obtained by different algorithms for analysis. Simulation test results demonstrate that the improved RBFNN exhibits lower prediction errors compared to its predecessor（with MSE of 0.000425 and MAPE below 10%）. Additionally,when compared to other machine learning algorithms and artificial neural networks,the proposed model exhibits superior performance in predicting the ship’s G_M.Therefore,the model proposed in this paper can serve as a tool for predicting the G_M of ships,providing accurate and real-time information for intelligent ship loading,and enhancing operational efficiency.