A forecasting methodology using support vector regression and dynamic feature selection

Various techniques have been proposed to forecast a given time series. Models from the ARIMA family have been successfully used, as well as regression approaches based on e.g. linear, non-linear regression, neural networks, and Support Vector Regression. What makes the difference in many real-world applications, however, is not the technique but an appropriate forecasting methodology. Here, we propose such a methodology for the regression-based forecasting approach. A hybrid system is presented that iteratively selects the most relevant features and constructs the regression model optimizing its parameters dynamically. We develop a particular technique for feature selection as well as for model construction. The methodology, however, is a generic one providing the opportunity to employ alternative approaches within our framework. The application to several time series underlines its usefulness.