Font Size: a A A

Research On Stock Price Prediction Based On Support Vector Machine

Posted on:2017-01-24Degree:MasterType:Thesis
Country:ChinaCandidate:F F ChenFull Text:PDF
GTID:2349330488965952Subject:Engineering
Abstract/Summary:PDF Full Text Request
The change of China's stock market is closely related to the market economy of the whole country,and it has an important influence on the growth of China's national economy.But there are also some weaknesses and problems in the stock market of our country,compared with the capitalist market,the stock market of our country only has 30 years of development,market structure is not mature,system is not perfect,supervision does not adapt,and too much short-term speculation,etc,so that the positive effect of the stock market is difficult to perform.These problems have led to a number of retail investors and corporate investors make mistakes in the investment,so the stock prediction and application research is very important and urgent to guide the reasonable investment of the majority of investors.Support vector machine(SVM)is a kind of new technology and new method of machine learning,VC dimension theory lays a solid theoretical foundation,and the structural risk minimization principle also provides a guarantee for its risk.Therefore,SVM in function fitting,high dimensional pattern recognition,time series prediction and other fields presented a unique advantage.The main research contents of this paper have the following several aspects:(1)Under the premise of consulting and referring to a large number of domestic and foreign literatures,not only for some of the stock prediction method to do a comprehensive elaboration,but also especially focus on explained support vector machines(SVM)and other related basic and theoretical knowledge.(2)Data selection and preprocessing.There are a lot of data in the stock market,and also have many kinds of stocks,involves different plate and the different industry categories,in this paper,we selected 60 stocks,the average distribution in the 20 industries,follow these stocks and download all kinds of index data needed in the experiment,after that,normalization processing these data.In this paper,we have carried out experiments on these industry stocks,and the selection of a few typical stock,and compared and discussed the results of the experiments.(3)Two typical kernel functions are studied in detail,the results show that the radial basis kernel function of the two kinds of kernel function is better.Secondly,the parameters optimization and selection of the radial basis kernel function are compared,through theexperiment,it is found that the effect of genetic algorithm optimization parameters will be better.(4)On the basis of the standard radial basis kernel regression SVM,by experimental comparison of different technical indicators:6 basic characteristic indexes,8 characteristic indexes(Contains the selection of indicators: MACD,RSI,OBV,CCI),comparative studies have found,the prediction accuracy of the 8 feature attributes is higher,and the error is lower.Then this paper makes a new attempt to predict the stock,by combining the application of fractal theory in the stock prediction Using power exponential distribution to solve the fractal dimension of the stock,experiment and compare the two different technical indicators with the trained model,from the results obtained,it is found that the fractal dimension is introduced to make the prediction result to be some extent improved.(5)At last,using the support vector machine and neural network,to compared the different indexes of the samples,you will find in different technical indicator system support vector machine prediction results are better than the neural network.Finally,the work of this dissertation made a comprehensive summary,as well as for further research to do guide.
Keywords/Search Tags:Stock Prediction, Support Vector Machine, Technical Indicator, Fractal Dimension
PDF Full Text Request
Related items