Non-stationary Time Series Prediction Research with Multi-factors Based on SVM

At present, only the single-factor is usually used in the process of non-stationary time series regression prediction based on Support Vector Machines thus inducing weak generative ability. To solve this problem, the concept of multi-factors is introduced in this paper. The problem of choosing the appropriate multi-factors used for elevating prediction ability is solved in this paper. And the paper proposes the method based on the Bayesian Networks structure model to obtain the dependency relationships among the factors, and moreover, determines factors set according to the dependency relationships in order to construct the Support Vector Machines regression model for time series prediction. The experiment results show that Mean Absolute Percentage Error is controlled below 10% through this method, and the prediction accuracy is improved compared with the value of support vector machines regression with the single-factor.