Multi-factors time series prediction based on PCV-SVM

Generalization performance of support vector machines (SVM) is affected by parameter selection. How to select optimal parameters to achieve the best training model has been a hot research spot. In order to improve generalization performance of SVM, K-fold cross validation is used to select parameters for training. However, K-fold cross validation is time-consuming, especially for large number of samples, and the method would require consumption of long time. In view of this problem, this paper raises parallel K-fold cross validation PCV algorithm which can decline runtime greatly. Meanwhile, we use this algorithm to select parameters of SVM. Finally, multi-factors time series regression prediction is experimented according to selected parameters. The experiment shows the PCV algorithm declines the runtime to about half of the existing ones on the condition of assuring accuracy.