Stock Prediction Research Based On Improved Dynamic Fuzzy Neural Network

In recent years,the constant deepening of national economic institutional reform and financial system perfection has made stock investment crucial in social activities.People are keen on the study of share price fluctuation and share price forecasting has grown to be a focus of massive investors.However,a deeper understanding would easily reveal that the stock market itself is a nonlinear system that hard to conceive and handle.Abundant factors internally interplay with each other while externally it is vulnerable to the distraction from outside environment,for which reason,the price fluctuation of stocks emerging to be sheer nonlinear and difficult to be precisely predicted.Secondly,the construction of stock forecasting system and the data scale required to be processed for prediction normally demand high for algorithm.It is precisely due to the complexity of nonlinearity that the expected outcomes of stock price forecasting model unable to be attained.Thus,the method of constructing a high-efficient and precise stock price forecasting model is of vital theoretical significance and practical application value to financial investors.This paper hereby studied in-depth the existing stock forecasting methods,in view of current problems and through dynamic integration of the outstanding attribute reduction ability of RS to remove redundant information of input data in the algorithm model and the applicability of DFNN(Dynamic Fuzzy Neural Network)to forecast and analyze the nonlinear system,a better appropriate stock forecasting model based on RSDFNN is proposed,whereby algorithm model structure is to be optimized based on forecasted outcomes.In case analysis phase,stock data of Suning,China Fortune Land Development Co.,Ltd.(CFLD)and other companies were collected to construct input data set.Features of data collected were then extracted and via optimized attribute reduction algorithm that based on genetic algorithm under rough sets theory,redundant data were reduced to remove.The obtained attribute was then employed as input of DFNN to trail predict and under results comparison and contrast of distinctive algorithm,effectiveness and accuracy of model building in the paper was then verified.Apart from considering deviation of share price numerical value upon setting of model performance index,this paper also incorporated the concept of prediction on directional symmetry and inflection point to verify predictability of the model towards tendency of share prices.For that purpose,in the data preprocessing,the share price trend was taken as the decision attribute to establish the decision table and then the core indicators of high correlation with it were selected through RS to improve the trend predicting accuracy in the follow-up prediction.Besides,a contrast verification of multiple conventional algorithms and the optimized forecasting model is set to show directly the effectiveness and practicability of the method mentioned in the paper in recommending stocks among multiple selection and reducing investment risks in real practice.