High Precision Stock Price Prediction Method Based On Data Perturbation Optimization

Stock price prediction is an important prerequisite for investors to obtain profits,and the accuracy of prediction directly affects the size of profits.In recent years,neural network based stock price forecasting methods have become mainstream forecasting methods and research hotspots.Previous research has shown that considering market public opinion factors to predict stock prices has higher accuracy.However,the difficulty in collecting market public opinion data is high,and the frequency of sudden public opinion events that have a significant impact on stock prices is low,making it difficult to meet the amount of data required for neural network training.At the same time,due to the black box characteristics of neural networks,it is difficult to understand the learned data characteristics and non-linear relationships,making it difficult to dynamically select the neural network with the highest prediction accuracy when data changes.This article conducts research on the above two issues,and the specific research content is as follows:(1)In order to reduce the amount of market public opinion data collected,a financial social public opinion dissemination model based on mutation mechanism was proposed to predict the development of public opinion.This method breaks the one-to-one transition rule of the SIR model and introduces a mutation machine to propose a one-to-many transition rule to describe the behavior choices of social platform users.Based on this,a susceptibility comment discussion immune model(SCDR model)is constructed to describe the participation of social platform users in public opinion dissemination,and to predict the development of public opinion through the changes in the number of various actors.In this paper,we use the simulation experiment to compare the changes of the model under different initial states and different propagation parameters,and use the Guba data to verify the model.Experiments have proved that the model effectively depicts the situation where public opinion causes users to comment or discuss,and describes the development of public opinion in detail and accurately,which can effectively predict the development of public opinion.(2)In order to consider the impact of sudden public opinion on stock prices and predict stock prices on the premise that the neural network training dataset does not contain public opinion information,a stock prediction method based on public opinion data perturbation was proposed.This method regards the impact of sudden public opinion on stock prices as a disturbance of public opinion to data,and uses the Shapley value to quantify the offset caused by the disturbance on stock price prediction results.When public opinion has an impact on stock prices,the magnitude of the disturbance caused by public opinion to the data is calculated based on the Shapley value and the development of public opinion.The disturbance is added to the original data to optimize the data,and the optimized data is used for stock price prediction.By comparing the prediction results of stock prices before and after data optimization using different neural network models,it is proved that this method can effectively improve the accuracy of stock price prediction.(3)In order to dynamically select the most accurate neural network from different neural networks when data changes,a stock price prediction method based on data change perturbation is proposed.This method treats changes in data as disturbances added to the original data,and measures the impact of data changes on the prediction results by calculating the Shapley value of the disturbance.When data changes,the Shapley value of the changed attribute is calculated to construct the basic probability assignment function of D-S evidence theory based on interval models.Then,cosine similarity is used to calculate the similarity of the evidence vector,and Shannon entropy is used to calculate the information content of the basic probability assignment function.The conflict basic probability assignment function is weighted based on the similarity and information content.Finally,the weighted conflict basic probability assignment function is used for fusion to obtain the final result,and the optimal neural network is selected for stock price prediction.The experimental results show that the prediction accuracy of this method is higher than using a single neural network model for prediction.