Prediction of Scour Depth Around Bridge Pier by Support Vector Machines

Scouring is one of the main causes of failures of bridges and piles in rivers and marine environment. So the estimation of scour depth around bridge piers and piles is of great importance. On the other hand, since the scour depth properties should be considered in designs by the designers, the importance of acceptable accuracy to estimate the scour depth properties will be quite highlighted. Regarding the importance of scouring investigation, there are several empirical formulas that have been presented by researchers but acceptable results have not been provided yet. Considering the fact that the prediction of scour depth around a pile is complicated and is affected by sediment characteristics and sediment transport mechanism, current properties and pile geometries, new approaches other than empirical ones are being sought by researches. Recently alternative methods like data mining approaches have been widely applied to simulate complicated problems. Artificial Neural Networks (ANN) as a famous data mining approach has been successfully used to estimate the scour properties around a pile. However, performances of Support Vector Machines (SVM) as another type of data mining approach are not explored yet. SVM has been recently applied in fields of particle identification, face identification, text categorization and bioinformatics. In this study SVM is applied to estimate the scour depth around a pile and the results are compared with those of the ANN by MLP network with one hidden layer and back propagation training algorithm. Performances of all methods are tested by experimental data sets and the results are compared using statistical measures. Results of statistical measures of verification stage indicate that SVM provides a better estimation of scour depth than ANN and empirical formulae. They also indicate that data mining approaches provide better prediction than empirical approaches.