Application of PCA-LSTM algorithm for financial market stock return prediction and optimization model

Abstract

Accurately predicting stock returns can help reduce market risk. This paper briefly introduced the long short-term memory (LSTM) algorithm model for predicting stock returns and combined it with principal component analysis (PCA) to improve the prediction accuracy. Simulation experiments were conducted on 80 stocks, and the PCA-LSTM model was compared with back-propagation neural network (BPNN) and LSTM models. The results showed that the PCA analysis method effectively identified the principal components of variable indicators. During the training iteration convergence, the PCA-LSTM model not only converged faster but also had smaller errors after stabilization. Moreover, the PCA-LSTM model had the highest prediction accuracy, the LSTM model was the second, and the BPNN model was the worst.