Abstract: Aiming to address the problems involving high complexity, cumbersome calculation processes, and insufficient accuracy in general vessel gross tonnage calculation, this study proposes a model approach that integrates feature engineering with multi-algorithm optimization. Initially, feature variables were extracted from principal ship dimension parameters, based on which three nonlinear regression models for gross tonnage were constructed to quantify the nonlinear tonnage relationships. Subsequently, regression prediction was performed separately using Backpropagation Neural Network (BPNN) and Random Forest (RF) algorithms, with model fitting and validation analysis conducted through nonlinear least squares and Particle Swarm Optimization (PSO) methods. Finally, experimental validation was conducted using a dataset of 1,913 fishing vessels in the South China Sea region, and the results demonstrated that both random forest and BP neural network algorithms delivered strong performance in fishing vessel gross tonnage regression prediction, with random forest exhibiting superior accuracy compared to the BP neural network approach. Among the optimization methods, the nonlinear least squares technique yielded better results than the particle swarm optimization algorithm across multiple evaluation metrics, particularly for mean bias error (MBE) , mean absolute error (MAE) and mean absolute percentage error (MAPE), as well as computational efficiency.Comparative analysis of the three regression models revealed that the hybrid gross tonnage prediction model consistently outperformed both the sub-index model and the linear product model across all evaluation indicators. This superior model achieved the following performance metrics: mean bias error (MBE) of -1.2444, root mean squared error (RMSE) of 32.0362, mean absolute error (MAE) of 24.481, mean absolute percentage error (MAPE) of 9.94%, and coefficient of determination (R²) of 0.9619, while also demonstrating excellent generalization capability and robust performance characteristics. Based on this, a comparative analysis was conducted among the actual ship’s gross tonnage, the International Gross Tonnage (GT) measurement formula, and the model predictions. The hybrid gross tonnage prediction model demonstrated higher accuracy than the international GT formula, validating the effectiveness of the model. This study employs artificial intelligence (AI) and machine learning (ML) methods to optimize the gross tonnage calculation model for fishing vessels, thereby improving accuracy and providing valuable insights for the digital and intelligent design and management of fishing fleets.