Research And Application Of Machine Learning In Stock Price Hybrid Prediction Model

With the development of financial markets,stock price forecasting is of great significance for making correct decisions and planning in the future.It can not only bring huge benefits to financial market investors,but also become a tool for risk management.The stock price sequence has the characteristics of non-linearity,non-stationarity and high noise.Using the new machine learning model to explore the stock price operation law and predict the stock price has practical and theoretical significance.The research shows that the traditional stock price prediction model is relatively simple,and only a single prediction method is used to train the original data.It is limited to the advantages of the basic prediction model,lacks the advantages of integrating multiple methods,and also lacks in error analysis.Therefore,this thesis proposes two new hybrid models based on data processing and model parameter optimization,which improves the accuracy and stability of stock price prediction.The main research work includes the following three aspects:(1)The theoretical basis and technical introduction of stock forecasting.The influencing factors of stock closing price and other neural network models used for comparison,autoregressive integrated moving average model(ARIMA),and variational modal decomposition(VMD)methods,optimization algorithms,and modeling processes involved in the research process are analyzed and introduced.At the same time,five model prediction performance evaluation indicators used in the study are introduced.(2)A hybrid model of gated recurrent cell network based on variational pattern decomposition and error correction is proposed.Firstly,the closing price of the stock is taken as the prediction object,and the VMD method is used as the preprocessing method to decompose the complex original data into several modal components at different frequencies.The gated recurrent unit(GRU)is used as the prediction model to predict the sub-series separately to obtain the preliminary prediction values.A VMD-ARIMA model is constructed to correct the error series after the initial forecasting was completed.Finally,the historical stock closing prices of CSI 300,S&P 500,Hengrui Pharmaceutical and CITIC Securities are used to verify that the model has good prediction performance.(3)A hybrid model for stock price prediction based on VMD and echo state networks(ESN)is proposed.The VMD method is used to decompose the original stock closing price series,Then,the genetic algorithm(GA)is used to obtain the optimal parameters of the echo state network(ESN),and the optimized ESN network is used to predict the decomposition subsequences respectively.The experiment selects the historical data of the stock closing price of CSI 300,S&P 500,and compares each single model and other hybrid models.The experimental results prove that the model improves the accuracy of stock price prediction and has strong stability.For the characteristics embodied in the stock price series,the two hybrid models based on machine learning proposed in this paper both perform well in prediction.In future research,the advantages of various data processing methods and machine learning models can be exploited to design hybrid models with more accuracy and stability.