Bridging Stock Price Data and Technical Features for Stock Forecasting: A Comparative Study of BiLSTM, CNN & GRU Models

This paper evaluates the effectiveness of financial indicators in deep learning models—specifically BiLSTM, CNN, and GRU—for stock price prediction. We conducted two experimental analyses using historical stock data: one based solely on raw closing prices and another incorporating technical indicators such as Simple Moving Average (SMA), Relative Strength Index (RSI), and Moving Average Convergence Divergence (MACD). Our comparative results demonstrate that including financial indicators significantly enhances predictive performance. The GRU model enhanced with financial indicators achieved an $\mathbf{R}^{\mathbf{2}}$ score of $\mathbf{0. 9 9 5 7}$, which is considerably higher than its version using only raw prices ($\mathbf{R}^{\mathbf{2}}=\mathbf{0. 8 2 1 8}$). Similarly, BiLSTM and CNN models showed substantial accuracy improvements with indicator integration. These findings underscore the importance of careful feature engineering, demonstrating that combining domain-specific indicators with deep learning techniques enhances the robustness and accuracy of stock market forecasting models, offering valuable insights for traders and investors seeking reliable predictions.