Enhancing Financial Time Series Predictions with a Hybrid BNN-LSTM Approach

Accurate forecasting of stock market indices is vital for guiding investment strategies and mitigating financial risks.This study proposes a novel hybrid Bayesian Neural Network-Long Short-Term Memory (BNN-LSTM) model to enhance the predictive accuracy of Dow Jones Industrial Average (DJIA) closing price forecasts.By integrating the uncertainty quantification capabilities of Bayesian Neural Networks with the sequential learning strengths of Long Short-Term Memory networks, the hybrid model addresses the challenges of modeling complex, nonlinear, and timedependent financial data.Comparative experiments were conducted using Bayesian Neural Networks, LSTM, Random Forest (RF), Gradient Boosting Machine (GBM), and the hybrid BNN-LSTM model on historical DJIA data spanning January 1, 2005, to December 31, 2022, for training, and January 1, 2023, to January 31, 2024, for testing.The hybrid BNN-LSTM consistently outperformed all competing models across multiple evaluation metrics.These results underscore the model's superior ability to capture complex market dynamics and its robustness in forecasting financial time series.This study contributes a powerful tool for financial decision-making and sets the foundation for future advancements in hybrid deep learning models for stock market analysis.