The school of Computer Science and Technology, Beijing University of Posts and Telecommunications, Beijing 100876, China

---

Abstract—We propose a novel approach of hyper-parameters selection for SVM regression when it is employed to make time series prediction. In this method, optimal hyper-parameters for SVM are obtained when the residual of training set follows white noise distribution. This conclusion is deduced from the fact that the targets of training set have inherent correlations with each other in time series which is different from other regression problems where the targets of training set are identically and independently distributed. Furthermore, by using this approach, confidence interval can be computed under any given confidence degree 1−α which is an important value for many applications. Two algorithms to compute confidence interval are listed in different cases. At last we compare the prediction results on two benchmark time series with cross validation method.

Index Terms—Support vector machines, hyper-parameter, time series prediction, white noise

[PDF]

Cite: Yanhua Yu, Meina Song, Junde Song, "A Novel Hyper-parameters Selection Approach for Support Vector Machines to Predict Time Series," Journal of Computers vol. 7, no. 12, pp. 2921-2930, 2012.