Abstract: Objectives In addressing the challenges of frequent parameter iterations and high computational costs associated with finite element analysis in ship structural design, this paper proposes a surrogate model based on machine learning methods for predicting the inherent vibration characteristics of double-layer cylindrical shells.Methods Taking the design parameters of a double-layer cylindrical shell as input features and its modal frequencies as the prediction target, data samples are generated through MATLAB-ANSYS co-simulation technology. Subsequently, four machine learning methods—Kriging, Support Vector Machine, Neural Networks, and Random Forest—are introduced to predict and analyze the inherent vibration characteristics of the double-layer cylindrical shell, followed by a comparative evaluation of their performance.Results The results indicate that all surrogate models achieved a coefficient of determination (R²) above 0.95. Among them, the relative errors between the predicted and true values for the Kriging, Support Vector Machine, and Random Forest methods were all controlled within 3%. Overall, the Kriging method demonstrated the best accuracy and robustness, making it highly suitable for predicting the inherent vibration characteristics of double-layer cylindrical shells.Conclusions The findings of this study can serve as a valuable reference for the design of underwater structures.