Bayesian networks, Gamma Test, Groundwater level, model Least Squares Support Vector Machine, Plain Ardebil

Groundwater has been raised as one of the major sources of water supply for drinking and agriculture, especially in arid and semi-arid. Simulation of groundwater system because of the complexity of these systems is a difficult task. In this paper, using data Ardabil plain water level in the period (1972-2011), the evaluation and selection of appropriate inputs for processing gamma test performance and efficiency of the least squares support vector machines and Bayesian network models were discussed. Monthly water level as input parameters with different delays Gamma test was considered. Gamma test results showed that the water level by 6 latency, offers better results to predict. Water level simulation using least squares support vector machines and Bayesian network models also showed that the input structure to predict the water level the next month will be delayed until six. The two models with the same input structure, least squares support vector machine model, better performance, according to the coefficient of determination 0.977, mean absolute error 0.204 and root mean square error 0.307, compared to Bayesian networks have. The results showed that gamma test compound in the appropriate input soft computing can have a better performance.