Study On Stock Time Series Based On Rough Set And Rbf Network

Traditional time series analysis has achieved good results in dealing with simple linear problems. But its strength does not match its ambition when facing complex nonlinear system such as stock. Although the emergence of artificial neural network provides a new effective way for the nonlinear time series analysis, there are still certain defects. Considering the difficulties that the neural network encountered in the stock forecasting, this study introduces Rough Set theory to the forecasting model, and launches deep research.(1) Reduce the stock data by using Rough Set theory. It can extract the core knowledge from original data and improve ananlysis efficiency. Considering that discretization based on conditional information entropy needs too much computing time, this study improves the process of discretization points selecting and computing. One heuristic algorithm for discretization based on information entropy is proposed and applied to the discretizing of the stock data.(2) Considering the weakness of GA in achieving effective attribute reduction, this study improves its genetic factors and reduces the samples by attribute reduction method based on improved GA, and gets better result.(3) Every learning algorithm of RBF network has its advantages and disadvantages. It's difficult to determine that which one is the best in practical application. This study compares three of the common algorithms through lots of experiments, and determines a suitable network model for the stock forecasting.(4) This study proposes the definition and classification of stock inflection points (three kinds of inflection points). A decision support model is proposed to forecast stock inflection points based on proper level data, and give a decision-making signal to assist the small investors in real-time operation.Finally, lots of experiments are made to prove that the method of stock time series analysis based on Rough Set and RBF network is efficiency in solving the problems of the network such as complex structure and slow learning speed.