Hybrid ARIMA-LSTM Model for Stock Market Prediction: A Time Series and Deep Learning Integration Approach

This study aims to evaluate the performance of the hybrid model based on ARIMA and LSTM in stock market forecasting and compare it with multiple traditional models to verify its superiority in dealing with complex nonlinear relationships and long-term dependencies. In terms of methodology, we preprocessed the raw data comprehensively. First, we used a time series-based interpolation method to fill missing values to ensure data integrity. Then, to make the data meet the model input requirements, all numerical data were normalized and scaled to the [0, 1] interval. In terms of data set division, the data was divided into training and test sets in a ratio of 80:20 to train and evaluate model performance. At the same time, we used correlation analysis and principal component analysis (PCA) for feature selection, retaining features that are highly correlated with stock market fluctuations, such as historical stock prices, trading volumes, GDP growth rates, inflation rates, etc., and PCA was used to reduce the dimension of features to reduce data redundancy. For the LSTM model, we constructed a network structure with 3 hidden layers. Each hidden layer contains 128 neurons, and ReLU is used as the activation function to enhance the nonlinear expression ability of the model. During training, the Adam optimizer was used, the learning rate was set to 0.001, and the batch size was 64. In addition, to prevent overfitting, a Dropout layer was added between the LSTM layers, and the Dropout rate was set to 0.2. In the result analysis, we used the Wilcoxon signed rank test to compare the results of the hybrid model with other traditional models to evaluate the statistical significance of the improvement. The results show that under the 95% confidence interval, the evaluation indicators (MSE, RMSE, R², MAE) of the hybrid model have significant advantages over the traditional model, further proving the effectiveness and reliability of the hybrid model in stock market forecasting.