Scour around pile groups is rather complicated and not yet fully understood due to the fact that it arises from the triple interaction of fluid–structure–seabed. In this study, two data mining approaches, i.e. Support Vector Machines (SVM) and Artificial Neural Networks (ANN), were applied to estimate the wave-induced scour depth around pile groups. To consider various arrangements of pile groups in the development of the models, datasets collected in the field and laboratory studies were used and arrangement parameters were considered in the models. Several non-dimensional controlling parameters, including the Keulegan–Carpenter number, pile Reynolds number, Shield's parameter, sediment number, gap to diameter ratio and number of piles were used as the inputs. Performances of the developed SVM and ANN models were compared with those of existing empirical methods. Results indicate that the data mining approaches used outperform empirical methods in terms of accuracy. They also indicate that SVM will provide a better estimation of scour depth than ANN (back-propagation/multi-layer perceptron). Sensitivity analysis was also carried out to investigate the relative importance of non-dimensional parameters. It was found that the Keulegan–Carpenter number and gap to diameter ratio have the greatest effect on the equilibrium scour depth around pile groups.
Keywords: ANN, pile group, scour depth, soft computing approaches, SVM