Prediction of scour depth around bridge abutments using ensemble machine learning models

Abutments are the structures that support the ends of a bridge deck. Scouring of streambed is a significant problem and ultimately results in the failure of the bridge when the abutments are exposed to flowing water over the long term. Abutment scour is influenced by the type of abutment, shape, and size of the abutments. In the current study, machine learning (ML) models have been utilized for predicting the scour depth around abutments making use of experimental data. The scour depth was modeled around three types of abutments: a vertical wall, a semicircular wall, and a 45° wing wall. Five input parameters, namely, the length of the abutment (L), breadth of the abutment (B), sediment size (d₅₀), approaching flow depth (h) and average approaching flow velocity (U), were used in this study. For predicting the abutment scour depth, ML models such as Adaptive Neuro-Fuzzy Inference System (ANFIS), Gradient Tree Boosting (GTB), Group Method of Data Handling (GMDH), and Multivariate Adaptive Regression Splines (MARS) were applied. Statistical metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Relative RMSE (RRMSE), Normalized Nash–Sutcliffe Efficiency (NNSE), Kling-Gupta Efficiency (KGE), and Willmott Index (WI) have been employed to evaluate the performance of each model. It was found that the GTB model provided relatively accurate predictions of the scour depth around the semicircular and 45° wing wall abutments with good metrics. Similarly, the MARS model outperformed all other models in terms of predicting vertical wall abutment scour depth.