Comparison of the support vector machine and radial function neural network models in predicting of SiminehRood river water quality Iran.

In this study, the performance of Support Vector Machine (SVM) and Radial Base Neural Network approach in predicting the water quality of SiminehRood River was examined. For this purpose, the Sodium Adsorption Ratio (SAR) and Chlorine ions were considered as indicators of water quality in agricultural use. Sodium, calcium, magnesium, pH, Ec, and river flow rate were utilized as input monthly parameters throughout a 12-year period (2003-2014). The results evaluated based on correlation coefficient, root means square error and mean absolute error. The results of the validation period in 4 stations of Pol Bukan, Dashband Bukan, Ghezel Gonbad and Kaullan showed that the SVM model in comparison with the neural network of the radial base function, has higher correlation coefficient (SVM: 0.71 to 0.94, RBF: 0.3 to 0.5), the lowest root means square error (SVM: 0.028 to 0.075 mg/l, RBF: 0.0672 to 0.317 mg/l), a lower absolute mean error (SVM: 0.003 to 0.033 mg/l, RBF: 0.087 to 0.19 mg/l) for the chlorine ion parameter and in the same order SVM values: 0.63 to 0.88 and RBF: 0.21 to 0.38, SVM: 0.0013 to 0.0282 mg/l and RBF: 0.047 to 0.025 mg/l, SVM: 0.0085 to 0.046 mg/L and RBF: 0.0653 to 0.0996 mg/l for sodium absorption ratio.Therefore, the Support Vector Machine model has better accuracy and performance for predicting water quality parameters of SiminehRood River than the Radial Basis Function Network.