Application of SVR and GRNN Models in Estimating Maximum Scour Depth at River Confluences under Live-Bed Condition

River confluences are one of the most complex places in river systems, that it is important to predict the maximum scour depth (Ds) at this place using intelligent systems that consider this complexity. Therefore, in this study, the performance of two artificial intelligence models, namely, SVR (considering different validation techniques including train-test, K-Fold and leave-one-out) and GRNN was evaluated. Results showed that, although all models show approximately good accuracy in predicting the Ds; but, SVR with train-test validation method shows more accuracy (with R2, MAE, MARE, RMSE and NSE of 95.66, 0.0124, 4.26, 0.0168 and 0.993, respectively), and after that SVR K-Fold (at K=9), SVR leave-one-out; and GRNN are the accurate models in this study, respectively.