Research On The Forecast Of Stock Price Fluctuation In A Stock Market Based On Data Mining Technology

With the development and progress of China's economy,more and more people pay attention to China's stock market.Stock investment has become an important way for investors to invest.There are many factors affecting the stock price fluctuation.If the change can be predicted successfully,it can effectively avoid the risk of stock investment and eliminate the malpractice of blind investment.Based on the construction of stock selection index,we use random forest algorithm and recursive feature elimination algorithm to filter out appropriate factor indicators from index pool to form index system.Then the stock returns are calculated according to the corresponding formula,and the results are divided into two states(rise and fall).The stock selection study is simplified to the two classification problem.Then use the sample data to adjust the parameters of processing of SVM model and BP neural network model,finally combining the SVM model and BP neural network model and AdaBoost model respectively,to construct the SVM_AdaBoost model and BP_AdaBoost model to predict the stock price,so as to help solve the problem of stock investment choice.This paper first introduces the background knowledge of A stock market,summarizes some research methods at home and abroad,analyzes several existing stock prediction model;then introduces the related knowledge of SVM model,AdaBoost model and BP neural network;then on the five aspects of the factors in the pool are described.The empirical part selects A shares as the research object,to verify the random forest algorithm and recursive feature elimination algorithm in feature selection on the feasibility,also verified the performance of RBFSVM prediction under different kernel functions and the performance of the BP neural network,empirical evidence shows that in addition to the poor performance of the sigmoid-RBFSVM algorithm,all other classifiers with good performance.However,the single SVM and BP neural network models are unstable.In order to improve the performance of the model,AdaBoost algorithm is introduced,and SVM_AdaBoost model and BP_AdaBoost model are constructed.The hybrid model has two model advantages: hybrid model can reduce the complexity of the SVM kernel parameter selection and BP neural network learning rate and hidden layer node selection problem,and the structural characteristics and parameters of SVM nuclear diversity and BP neural network learning rate and initial solution makes the SVM model and BP neural network model can training a variety of weak classifiers in a variety of values;and finally through the AdaBoost can be combined with characteristics of constructing multiple classifier and fault classification samples out of concern for a strong classifier,to predict the stock price change.The empirical results show that the SVM_AdaBoost model and BP_AdaBoost model have a certain degree of improvement in prediction accuracy compared with the original SVM model and BP neural network model,and the SVM_AdaBoost model is the best.However,the above models are unstable.Therefore,we need to further study this problem in the future,making the SVM_Ada Boost model and BP_AdaBoost model get better results in the prediction of stock price fluctuation.