Predicting the Longitudinal Dispersion Coefficient Using Support Vector Machine and Adaptive Neuro-Fuzzy Inference System Techniques

Much research is carried out for predicting the longitudinal dispersion coefficient (LDC) in natural streams based on regression models. However, few methods are accurate enough to predict the LDC parameter satisfactorily. In the present investigation, two data-driven methods for predicting the longitudinal dispersion coefficient are developed based on the hydraulic and geometric data that is easily obtained in natural streams. We have tried to determine the deficiencies of previously developed longitudinal dispersion models, and subsequently develop an optimum model. For this purpose, a support vector machine (SVM) that is based on structural risk minimization and adaptive neuro-fuzzy inference system (ANFIS) models have been used, and the results are compared. Findings indicated that the newly developed models are considerably better than previously developed models based on classical regression techniques. This article shows that SVM and ANFIS models predict the LDC with a correlation coefficient (R) greater than 0.70 (R = 0.73 and 0.71, respectively). Furthermore, the results obtained using the SVM based on threshold statistic analysis are better than the ANFIS model. In other words, the SVM model has a less error distribution in testing step than the ANFIS model.