Application Of Boosting And Bagging Algorithms In Predicting Bridge Scour Under Clear-Water Conditions

Abstract

Accurate prediction of scour depth around bridge abutments is critical for ensuring the safety and longevity of hydraulic structures, especially under clear-water conditions. This study investigated the predictive performance of two ensemble machine learning models including Extreme Gradient Boosting (XGBoost) and Random Forests (RF) to estimate scour depth for three distinct abutment geometries: Vertical Wall, 45° Wing Wall, and Semicircular Abutments. A laboratory-generated experimental dataset was employed with a 70:30 split for training and testing respectively. The results demonstrated that XGBoost consistently outperformed RF across all abutment types achieving higher determination coefficients (R²) and lower error metrics (Root Mean Square Error and Mean Absolute Error). It was observed that XGBoost achieved an R² of 0.9707 for Vertical Wall abutments, compared to 0.8721 by RF. The superior performance of XGBoost is attributed to its gradient-boosting framework and regularization capabilities, which enhance its generalization ability on complex, nonlinear datasets. This study confirms the effectiveness of XGBoost as a reliable and accurate tool for predicting scour depth, outperforming traditional ensemble methods. The findings highlight the potential of advanced machine learning approaches in improving hydraulic design and risk assessment practices.