Evaluation of the Efficiency of Support Vector Regression, Multi-Layer Perceptron Neural Network and Multivariate Linear Regression on Groundwater Level Prediction (Case Study: Shahrekord Plain)

Accurate and reliable simulation and prediction of the groundwater level variation is significant and essential in water resources management of a basin. Models such as ANNs and Support Vector Regression (SVR) have proved to be effective in modeling nonlinear function with a greater degree of accuracy. In this respect, an attempt is made to predict monthly groundwater level fluctuations using Multivariate Linear Regression, Multi-Layer Perceptron neural network models and two SVRs with RBF and linear function. In the present study, monthly data (from 2000 to 2010) of 18 observational wells in Shahrekord Plain were used for simulating and predicting the groundwater level. Regarding to NS efficiency and RMSE criteria, MLP model in 56% and SVR in 44%, have the best performance in comparison with other models. For an instance, in well No. 1, SVR-RBF using input parameters of groundwater level, temperature, evaporation and precipitation is superior to other models. General efficiency of MLP, SVR-RBF, and SVR-Linear for NS criteria is 0.703, 0.656 and 0.655, respectively; and for RMSE criteria is 0.857, 0.905 and 0.914 meter, respectively. Results indicate that MLP and SVR models give better accuracy in predicting groundwater levels in the study area when compared to the linear model.