Abstract: Objectives Currently, approximation techniques have been widely adopted in ship hull optimization design to mitigate the high cost and long cycle times associated with high-precision numerical computations. However, existing approximation models typically rely on predefined parametric modeling methods and manually defined parameters, which makes it difficult to uniformly represent hull geometry and poses challenges for subsequent dataset expansion and the development of general-purpose predictive models. Methods To address these issues, this paper proposes a rapid prediction model for wave resistance coefficients based on graph neural networks, named PB-GraphPR, which takes three-dimensional hull point cloud data as input. Built upon the Dynamic Graph Convolutional Neural Network (DGCNN), the model incorporates a multi-neighborhood parallel feature extraction structure to enable complementary learning and effective integration of local geometric details and global hull line features.Results The results show that the PB-GraphPR model can accurately and rapidly predict the wave resistance coefficient of the Series 60 standard model, achieving a coefficient of determination R²of up to 98.9% on the test set. Conclusions Furthermore, gradient-weighted heatmaps are employed to visualize the regions of focus of the model. This research provides new insights and methods for rapid ship performance prediction and serves as a reference for subsequent data-driven ship design studies.