Research Of The Stock Data Based On Machine Learning Methods

The stock has been developing for more than 400 years. The change in the stock market, reflect the name of economic development throughout the country situation and the operations of the stock company. The stock market is the entire social economy "barometer" and "alarm". Research of the changes of stock market data in the next few days is very important for the majority of investors.The stock data researchers can be divided into three categories. The first class of scholars applies technical analysis methods to study the behavior of the market in the past and present, to predict a general trend of the future stock market and future stock price. The second class of scholars applies time series methods to predict stock market movements. The third class of scholars applies machine learning methods to establish some related learning models to study the stock data.By introducing current research on the prediction of stock, machine learning, statistical learning theory, SVM (Support Vector Machines), chaotic time series, wavelet theory, this study built the prediction model based on these theories and the prediction model core was the learning methods of SVM. SVM is based on the statistical learning theory. It has a simple structure, and is good at solving the small sample and nonlinear problems, and can avoid the phenomenon of "dimension disaster" and "over learning".This paper has completed the major work as follows:Firstly, the prediction model was built based on the methods of support vector machines and chaotic time series. Specifically, according to the method of phase space reconstruction, the stock data was changed into time series data, and this time series data can be further divided into the training data and the testing data. Then, a prediction model was built based on SVM.Secondly, genetic algorithm was used to search the optimal parameters. The searching of the parameters of SVM directly affects the quality of prediction model. As genetic algorithm has the excellent ability of solving the problems of nonlinear and optimization in multi-dimensional space, application of parameters obtained from genetic algorithm can get better prediction result in this paper.Thirdly, wavelet transform was applied to change the stock data into high frequency data and low frequency data, then the prediction results of high frequency data and low frequency data was calculated by using the previous prediction model. Finally, the last prediction result was obtained by combining two kinds of prediction results based on wavelet reconstruction. Although using this method better result was achieved, but the prediction effect was associated directly with the selection of the wavelet function.