Study On Stock Price Prediction Based On Optimized Support Vector Regression

Nowadays, people are more and more interested in stock exchange. They hope to benefit from the transaction by analyzing a variety of information in stock market which can be used to instruct the investment.However, stock prices are infected by various factors. Therefore they have the characteristics of nonlinearity, time-variation and highly unstabilitywhich make stock prices very difficult to predict. Traditional approaches such as time series analysis can't extract the nonlinear features in stock data,andneural network algorithm is easily fall into over-fitting and local minimum points. Furthermore,stock data contain so much noise and redundant information, which limitthe prediction accuracy.As to these questions, this study cites linear local tangent space alignment from manifold learning, utilizes optimized support vector regression to predict the stock price. The main contributions of this thesis are as follows:(1)To summarize and review the present state of the domestic and foreign researches in stock price prediction, note the shortcomings of those proposed approaches.(2)To introduce knowledge and principle on stock price prediction, as well asalgorithm theory of support vector regression which would apply on stock price forecasting.(3)To cite linear local tangent space alignment from manifold learning for the first time in the field of stock price prediction, extract the pattern contained in stock data which is seen as a nonlinear manifold.It can reduce the noise, prune the redundant and improve the discriminability of extracted features.Finally the prediction accuracy is enhanced.(4)To propose a novel stock price prediction model which integrates linear local tangent space alignment and optimized support vector regression. Firstly, linear local tangent space alignment algorithm is used to extract features of stock raw data. Then use support vector regression model the nonlinear relationship between those features and stock close price, meanwhile use genetic algorithm to optimize parameters of support vector regression.Finally, the prediction accuracy is improved.(5)To validate adaptability and generalization of the model proposed in this study, the stock data from different country, category and time which include Shangzheng index, SP&500 index, IBM, Microsoft is collected. To compare with 4 traditional feature extracted approaches, and 3 classical stock price predicted approaches.The researchshows the proposed model outperforms others and has higher prediction accuracy and more powerful generalization ability. It has prodigious application value and can be used to guide stock transaction.