Modelling monthly runoff by using data mining methods based on attribute selection algorithms

Given the importance of catchment basin output flow for surface water management, precise understanding of the relationship between the amount of runoff and climatic parameters such as precipitation and temperature is important. therefore the identification of parameters are important in the modeling process. In this paper, after homogeneity tests have been carried out for monthly precipitation, temperature, and runoff data in the Navroud Catchment Basin in Iran, two combinations of effective factors for runoff are considered according to Relief and Correlation algorithms. A new Relief Algorithm first identifies effective features within a set of data in an orderly manner especially when the amount of available data is low. The new method uses a data-related weight vector average and a threshold value. Applying support vector regression and the nearest neighbor method, monthly runoff was modeled based on the two proposed combinations. The results showed that support vector regression approach which utilizes a radial basis function kernel, yields higher accuracy and lower error than the nearest neighbor method for estimating runoff. The improvement is particularly noticeable for flooding situations.