Abstract
Monopile foundations are widely used in marine engineering, but local scouring around the monopile can significantly reduce its bearing capacity. Accurately predicting the equilibrium scour depth is essential for the design of marine structures. This study employs Bayesian optimization to tune hyperparameters for machine learning models, such as Support Vector Machine (SVM), Random Forest (RF), Adaptive Boosting (AdaBoost), and eXtreme Gradient Boosting (XGBoost), to establish a comprehensive prediction model for scour depth around monopiles. Results demonstrate that Bayesian optimization effectively regulates model hyperparameters, enhancing model accuracy while reducing overfitting risks. The optimized machine learning models precisely predict scour depth under both wave-only and combined wave-current conditions. Performance comparisons show that all three ensemble learning models, both before and after optimization, outperform the standalone SVM model. Across all optimized models, the BO-AdaBoost model achieved the optimal Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R2) values during both training and testing phases, exhibiting the strongest model robustness. Furthermore, its predictive accuracy significantly surpasses that of existing empirical equations. Sensitivity analysis reveals near-bed velocity induced by current (Uc) as the most influential parameter, as changes in Uc alter sediment shear stress and subsequently affect scour depth.