Multivariate Time Series Prediction Based on Multi-Output Support Vector Regression
An improved support vector regression (SVR) model is presented in this paper and the model can train and predict both the multiple input and output samples, which can avoid SVR’s modeling for each output individually when predicting the multivariate time series. The proposed multi-output SVR (MOSVR) model can guarantee the regression ability of each output by choosing different kernel functions and model parameters for different outputs of one single optimization problem. On this basis, the norm summation of regression weight vector, the error summation of each output and the total error are minimized so as to be sure that the MOSVR model satisfies the structure risk minimization (SRM) principle. The experimental results based on several multivariable time series show that the MOSVR has better adaptability and gains less total regression error than the SVR.
KeywordsSupport vector regression Multi-output SVR Multivariate time series Time series prediction
This work was jointly supported by the National Natural Science Foundation for Young Scientists of China (Grant No: 61202332) and China Postdoctoral Science Foundation (Grant No: 2012M521905).
- 2.Han M, Fan MM (2006) Application of neural on multivariate time series modeling and prediction. In: Proceedings of the 2006 American control conference, vol 7. Minneapolis, Minnesota, USA, pp 3698–3703Google Scholar
- 3.Tsang, S., Kao, B., Yip, K.Y.: Decision trees for uncertain data. In: Proceedings of the 2009 IEEE international conference on data engineering. Washington, DC: IEEE Computer Society, pp 441–444Google Scholar
- 4.Yang K, Shahabi C (2004) A PCA-based similarity measure for multivariate time series. In: Proceedings of the 2nd ACM international workshop on multimedia databases. New York, ACM Press, pp 65–74Google Scholar
- 9.Guan HS, Jiang QS (2009) Pattern matching method based on point distribution for multivariate time series. J Softw 20(1): 67–79 (In Chinese)Google Scholar
- 12.Tang FM (2005) Research on support vector machine algorithms based on statistical learning theory. PhD Dissertation, Huazhong University of Science and TechnologyGoogle Scholar
- 13.Bi JB (2003) Support vector regression with application in automated drug discovery. Rensselaer Polytechnic Institute, Troy, New York, pp 87–92Google Scholar
- 14.Ren R, Xu J, Zhu SH (2006) Prediction of chaotic time sequence using least squares support vector domain. Acta Physica Sinica 55(2): 555–563 (In Chinese)Google Scholar