Predicting the stock prices of listed companies based on the LSTM model of recurrent neural network

Abstract

This experiment uses the LSTM model and establishes a two-layer LSTM structure to enhance the learning ability of long-term dependencies. The Dropout layer can prevent the model from overfitting historical data during training. Normalization, time series sliding window technology, and 80-20 training-test split method are used for data preprocessing. MAE, RMSE, and MAPE are used for evaluation to compare the predicted value with the actual value.