How Reliable Are ANN, ANFIS, and SVM Techniques for Predicting Longitudinal Dispersion Coefficient in Natural Rivers?

Determination of longitudinal dispersion coefficient (LDC) using artificial intelligence (AI) techniques can improve environmental management strategies for river systems. However, the uncertainty involved in AI models has rarely been reported. The main objective of this paper was to investigate the reliability of three AI-based techniques, including the artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS), and support vector machine (SVM), for predicting the LDC in natural rivers. To that end, LDC predictions were first carried out using ANN, ANFIS, and SVM techniques. Then, a forward selection (FS) and gamma test (GT) were conducted to sort input variables according to their importance and effects on LDC prediction. Finally, uncertainties in the model predictions were analyzed to answer the question, “How reliable are ANN, ANFIS, and SVM techniques?” It was found that model inputs could not be satisfactorily sorted by a linear method (i.e., FS) due to the complex and nonlinear nature of LDC. Thus, the nonlinear GT technique was chosen as a suitable input selection method for prediction of LDC. The results or model input variables selected from the GT technique showed good consistency with previous researches. Furthermore, the reliability of ANN, ANFIS, and SVM models was calculated and tabulated by an uncertainty estimation for LDC prediction. A high uncertainty was found in the models although they predicted LDC appropriately. It was also found that the uncertainty in the SVM model was less than those in the ANN and ANFIS models for estimating the LDC in natural rivers. The ANFIS model performs better than the ANN model.