Comparing the Performance of Support Vector machines and Bayesian Networks in predicting daily river flow (case study: Barandoozchay River)

Accurate estimation of river discharge is one of the important steps to optimum use of water resources. In this study, Support Vector Machines (SVM) and Bayesian Networks (BNs) methods were used to forecast daily discharge of Barandoozchay River. The daily discharge data of Barandoozchay River measured at the Dizaj hydrometric station during 2006 to 2010 was used for modeling, which 80% of the data used for training and remaining 20% used for testing of models. For assessing the role of memory in increasing or reducing of model accuracy, we tested different combinations of input variables. The results showed that at first, the accuracy of models increased with increasing of memory, as the most accuracy obtained in third combination of input variables in both of the methods. After that, with increasing of memory, the accuracy of models decreased. Comparing the performance of SVM and BNs models indicated that the accuracy of the SVM method with the R=0.976 and RMSE=1.80 (m3/s) was slightly more than BNs method with R=0.964 and RMSE=1.96 (m3/s).