Font Size: a A A

The Predict Of Stock Index Time Series Based On Support Vector Machine

Posted on:2009-08-08Degree:MasterType:Thesis
Country:ChinaCandidate:K XuFull Text:PDF
GTID:2189360278959234Subject:Management Science and Engineering
Abstract/Summary:PDF Full Text Request
The analysis and prediction of stock market is a very complicated system with small samples. This article introduces the statistic study theory and makes some research on the stock market by combining chaos analysis method. In the practical application, We choose Lssvm.m,a kind of anamorphous of SVM.. In order to ensure the preciseness of the data handling, this article do some validate on the effect of fitting and predict. It was found that the effect of predict is not well. We improve it by detailed analysis. At last, the improved Lssvm.m get better effect.The main research contents and conclusions are below. Advantages on processing limited sample and mathematical model of support vector machine are analyzed, and the disadvantage on practice apply is introduced. A kind of tool LSSVM.M which bases on the stared SVM's evolutionary method—least squares SVM (LS-SVM) is set up to process the time series. After analyzing the fitting and forecasting effect of LSSVM.M, it concludes that parameters selection is the main factor of affecting the LSSVM.M's assessment. In order to improve the tool's performance, this paper analyzes the theory of grid search and leave-one-out validation and sets up an advanced parameters selected method. That is, before optimizing the parameters, the parameters scopes will be selected. At last, the advanced tool's effect is verified with an example. Finally, the advanced LSSVM.M tool is used to forecast the performance of time series, In conclusion, this article introduce a method to analyze time series and discuss how to optimize the parameter of LSSVM and use it to predict the stock index effectively.
Keywords/Search Tags:Statistical learning theory, Support vector machine (SVM), Parameters optimization, Time series
PDF Full Text Request
Related items