为了估算中国东部沿海未扰动海相土的最大动剪切模量，建立了一个可靠的机器学习（machine learning, ML）预测模型。首先，构建了一个中国东部沿海未扰动海相土最大动剪切模量值的数据库。然后，应用了4种不同的ML模型，并对其预测精度进行了比较，以确定表现最佳的模型。最后，为了进一步提高预测的可解释性，采用了SHAP（shapley additive explanations）法来评估各个输入参数对最大动剪切模量的影响。研究结果表明，梯度提升（gradient boosting, GB）模型在预测准确性和对新数据的泛化能力方面优于其他ML模型。深度对海洋土壤动剪切模量的影响最为显著，其次是塑性指数。

To estimate the maximum dynamic shear modulus of undisturbed marine soil on the east coast of China, a reliable machine learning (ML) prediction model was established. First, a database of maximum dynamic shear modulus for undisturbed marine soils along the eastern coast of China was constructed. Then, four different ML models were applied, and their prediction accuracy was compared to determine the best model. Finally, in order to further improve the interpretability of the predictions, the SHAP(shapley additive explanations) method was used to evaluate the impact of each input parameter on maximum dynamic shear modulus. The results show that the gradient boosting (GB) model outperforms the other ML models in terms of prediction accuracy and generalization ability for new data. The depth has the most significant effect on the dynamic shear modulus of marine soil, followed by the plasticity index.