Ensemble SVR for prediction of time series
Yufeng Deng, Xing Jin, Yixin Zhong · 2005
Recently, support vector machine (SVM) as a new kernel learning algorithm has successfully been used in nonlinear time series prediction. To improve the prediction performance of SVM, We concentrate on ensemble method. Bagging and boosting, two famous ensemble methods, are examined in this paper. Experiments on two data sets (sunspots and Mackey-Glass) have shown that bagging SVR and boosting SVR could all improve the performance when compared with single SVR. For boosting, weighted median is a better choice for combining the regressors than the weighted mean.