• The ensemble learning-based shear wave speed model significantly improves prediction accuracy compared to conventional empirical formulas. • The model demonstrates robust generalization capability across different marine regions. • Application of model interpretability techniques reveals that the average grain size is the key dominant controlling factor. Shear wave speed of seafloor surface sediment is one of the essential parameters for marine engineering survey and also serves as an important parameter for marine geological exploration, resource assessment, and geological disaster prediction. Constructing a high-precision predictive model for shear wave speed is a crucial means of obtaining this parameter. In this study, a large number of sediment core samples were collected from the East China Sea (ECS) continental shelf, and their shear wave speed, density, water content, porosity ratio, and grain composition were measured in the laboratory. A machine learning prediction model for the shear wave speed of seafloor sediments was established using the LightGBM (Light Gradient Boosting Machine) algorithm, and the impact of multiple physical parameters on the prediction results was analyzed through model interpretability techniques. The results indicate that compared to traditional single and dual-parameter predictive equations, the LightGBM predictive model reduced the Mean Absolute Error (MAE) by 1.42–2.82 m/s and Mean Absolute Percentage Error (MAPE) by 4.13–9.94%, effectively improving the prediction accuracy of shear wave speed of seafloor sediments. Meanwhile, it demonstrated strong extrapolation and generalization capabilities in cross-sea-region prediction experiments. This high-precision interpretable model can provide strong support for actual marine engineering measurements.
Chen et al. (Sat,) studied this question.