Stock Price Prediction Based on the Fusion of CNN-GRU Combined Neural Network and Attention Mechanism

Accurately predicting stock prices can help investors make wise decisions, avoid blind herd trading behaviors such as chasing price increases and killing drops, and significantly impact the healthy development and the safe operation of the stock market. Stock time series data is highly complex and has huge data. This paper proposes a CNN-GRU-attention model for predicting stock prices, which uses three different data decomposition methods, including EMD, EEMD, and CEEMDAN for data preprocessing and selects the optimal data preprocessing method for the model. Furthermore, the integrated model is compared against CNN-LSTM-attention, GRU, RNN, and other conventional single models. Through the analysis of evaluation indicators such as MAE, RMSE, and R, it was found that the CNN-GRU-attention model had the best prediction accuracy. The experimental results of the dataset show that the CNN-GRU-attention model is feasible and universal in its prediction effect, with predicted values closest to the actual values and are better enough to meet the application’s needs.