Stock Data Analysis Based On BP Neural Network Multi-Combination Model

With the vigorous development of our country’s economy,people’s investment awareness has gradually enhanced.More and more people start going to the stock market,bringing a large number of financial trading volume and generating a lot of data.However,the traditional research methods of the stock market are difficult to deal with such complex data.Therefore,researchers take advantage of the excellent nonlinear fitting ability of neural network to process a large amount of data to conduct financial data research.So far,academia is still exploring how to choose appropriate input variables.In the selection of variables,this paper mainly optimizes the following four indexes in the stock market:market trend index,market energy index,volatility index and sentiment index^[1].These four categories basically cover the types of factors that affect stock market volatility,which is the variable that researchers are most concerned about.Under each large index,the representative small index is further selected,and the optimal input variables are selected through the different models constructed.Genetic algorithm selects excellent genes based on the concept of"survival of the fittest",so as to continue the next generation.The problem of input variable selection can be solved effectively by using this idea in neural network input variable selection.In fact,individuals with fewer input variables perform better when prediction errors are close.Based on this,the paper proposes a new fitness function combining two factors,which can not only ensure a small number of variables,but also have a better prediction effect.This paper selects the CSI 300 index as the research object,because its constituent stocks have strict selection criteria,the index has a high market coverage rate and the industry is evenly distributed,so it can basically reflect the stock market fluctuations.A total of 3600 groups of data from March 2006 to March 2021 are selected to set training group and test group.Based on the learning of BP neural network,the article firstly introduces the advantages and disadvantages of each research method by citing PCA analysis method and GA genetic algorithm.On this basis,the multi-combination prediction model based on BP neural network model is constructed,and the fitness function of GA genetic algorithm is improved.Finally,the prediction performance of different models is obtained through experiments.The experimental results show that the prediction effect of the multi-combination prediction model is better than that of the single BP neural network prediction model.The prediction results obtained by PCA analysis are improved by 1%compared with the unimproved GA genetic algorithm;Compared with the unimproved GA-BP neural network prediction model,the prediction accuracy of the improved GA-BP neural network prediction model is improved,and the efficiency of optimizing the number of variables is increased by about 55%,which provides a new idea for the selection of input variables in the stock market.