Hydrologic data exploration and river flow forecasting using self-organizing map and support vector regression

The support vector regression (SVR) is a novel and robust machine learning approach that has been successfully applied to solve problems related to river flow forecasting. However, an important step in the SVR modeling methodology that has received little attention is the selection of appropriate model inputs. This paper presents an input determination method that can be used to select significant input variables to a SVR based river flow forecasting model. The forecasting model was used for modeling daily river flows in a humid basin with seasonal rainfall pattern. The input determination method was developed using self-organizing map (SOM). The SOM was utilized to explore all the potential model input variables and then to identify inputs that have a significant relationship with the output variable. Consequently, the knowledge extracted from the input determination process was used to improve SVR model performance. Empirical results indicated that input determination based on SOM was helpful for developing logically sound SVR forecasting models.