Approximation in scour depth around spur dikes using novel hybrid ensemble data-driven model

The scouring process near spur dikes poses a threat to riverbank stability, making it crucial for river engineering to accurately calculate the maximum scour depth. However, determining the maximum scour depth has been challenging due to the intricacy of scour phenomena surrounding these structures. This research introduces a reliable ensemble data-driven model by hybridizing random tree (RT) using additive regression (AR), bagging (B), and random subspace (RSS) for predicting scour depths around spur dikes. A database of 154 experimental observations was collected from literature, with 103 and 51 observations used for training and testing subsets, respectively. A dimensionless analysis was performed on the collected dataset, selecting four variables as input variables (v/v_s, y/l, l/d₅₀, and Fd₅₀) and d_s/l as response variables. The performance comparison demonstrates that B_AR_RT has a better coefficient of determination (R²) of 0.9693, root mean square error (RMSE) of 0.1305, and Nash-Sutcliffe efficiency (NSE) of 0.9692. Finally, a comparison of the best hybrid model has been done with previous studies, and sensitivity analysis is performed to determine the most influential parameter for predicting the scour depth around spur dikes.