A Stock Prediction System Based On Data Mining

With the development of economy and society,more and more citizens in China tend to invest in the stock market to achieve appreciation in asset value.It is a hard task to predict the ups and downs of stocks and choose the high quality ones from thousands of stocks.Data mining,which could explore hidden information by algorithm from a great mass of data,is widely used in the stock analysis.This thesis builds the model for the change of stock price sequence,using stock indicators as the analysis foundation and Support Vector Machine(SVM)as the algorithm,aiming at forecasting the fluctuation of stocks by the method of machine learning and data mining.A feature construction method based on technical indicator sequence is proposed.Additionally,according to the thought of maximization of profit,this article presents the modified Support Vector Machine using Genetic Algorithm(GA)for parameter optimization based on AUC value-the area under the ROC curve,to improve the low usability in traditional methods for prediction of stocks.In the experiment,it shows good classification performance and generous profits on the prediction task for the fluctuation of Shanghai Pudong Development Bank(SPDB)stock,by application of the algorithm and optimization of the result on different time windows to seek for the best buy points.Then,this article implements a program used for quantitative stock selection.The algorithm would dig for regularity of rising for a great number of stocks,and choose to invest in the stock which has the maximal probability.By simulating the investment decision-making procedure and computing the earning of the strategy,the result of the experiment indicates clearly that,the prediction method proposed by this thesis,to some extent,could direct the stock ransactions.Therefore,it could help investors simplify analysis steps and increase investment income.At last,based on the algorithms and experiments above,a stock prediction system is implemented in MATLAB.This system includes four modules--stock data obtainment,stock market visualization,stock trading points prediction,and quantitative stock selection-which offers convenience to do stock analysis for standard investors.