Abstract:In order to improve learning efficiency and generalization ability of SVM, firstly, the raw data is preprocessed by using an information granulation algorithm. This algorithm can divide sample space into multiple particles (subspace), reduce the sample size and save the time complexity. And then, the granulated information take SVM to carry on the regression analysis, while take cross validation to select the optimal classifier adjustable parameters, which can reduce the complexity of the classifier and improve the generalization capability of the classifier and avoid Over learning and less learning. Finally, the test results on forecasting the Shanghai Composite Index have proved that the system has a good-performance and make time series regression prediction precisely.