Research on financial data trend prediction based on deep learning

The complex market behavior mechanism makes the financial time series have the characteristics of noise, low signal-to-noise ratio, non-linearity and non-stationarity, and fluctuation agglomeration, which makes the traditional econometric forecasting model have great challenges in predicting the trend of financial data. It is of great practical significance and theoretical value to use neural network algorithm and deep learning technology to predict the time series of financial market. Therefore, based on the deep learning model LSTM (Long Short-Term Memory), this paper introduces the VAT (Virtual Adversarial Training) regularization method which can increase the robustness of the model, and proposes the VAT-LSTM financial data trend prediction model to solve the problems of nonlinear distribution characteristics and generalization error in financial data. Through the empirical analysis of the Shanghai Composite Index data from 2015 to 2022, it is found that the model performs well in short-term, medium-term and long-term prediction. Compared with the classical time series prediction models such as LSTM, RNN, GRU, BP, SVR and ARIMA, it has higher prediction accuracy. It is suitable as a prediction model for financial data trends and can be extended to other financial research fields.