A Fast Grid Search Method in Support Vector Regression Forecasting Time Series

Selection of kernel function parameters is one of the key problems in support vector regression(SVR) for forecasting because these free parameters have significant impact on the performances of forecasting accuracy. The commonly used grid search method is intractable and computational expensive. In this paper, a fast grid search method is proposed for tuning multiple parameters for SVR with RBF kernel for time series forecasting. Empirical results confirm the feasibility and validation of the proposed method.