A novel boosting ensemble committee-based model for local scour depth around non-uniformly spaced pile groups

Prediction of the scour depth around non-uniformly spaced pile groups (PGs) is one of the most complex problems is hydraulic engineering. Different types of empirical methods have been developed for estimating the scour depth around the PGs. However, the majority of the existing methods are based on simple regression methods and have serious limitations in modelling the highly nonlinear and complex relationships between the scour depth and its influential variables, especially for the non-uniformly spaced pile. Hence, this study combines prediction powers of tree popular machine learning (ML) methods, namely, Gaussian process regression (GPR), random forest (RF), and M5 model tree (M5Tree) using novel Least Least-squares (LS) Boosting Ensemble committee-based data intelligent technique to more accurately estimate local scour depth around non-uniformly spaced pile groups. A total of 353 laboratory experiments data were compiled from published papers. non-dimensional results obtained demonstrated that the ensemble model can more accurately estimate the scour depth than the individual predictions of the GPR, RF, and M5Tree models. The proposed Ensemble model with correlation coefficient (R), root mean square error (RMSE) and mean absolute percentage of error (MAPE) of 0.972, 0.0153 m, and 10.89%, respectively, significantly outperformed all existing empirical models. Furthermore, the sensitivity analysis demonstrated that the pile diameter is the most influential variable in estimating the scour depth.