Stock price prediction is extremely important in financial analysis, especially for volatile stocks like Tesla's. However, accurate prediction remains challenging due to the non-linear, dynamic nature of stock market data. Although deep learning algorithms such as Deep Unfolding (DUF), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN) have enhanced the capacity for observing intricate temporal and spatial patterns, individual or sequential applications of these algorithms often fail to capture dependencies between time steps or prioritize the most important input features, thereby limiting prediction accuracy. To address these limitations, a Hybrid CNN-LSTM with Attention (HCLA) model is suggested. This model utilizes CNN for spatial features, LSTM for temporal learning, and an attention mechanism to focus on significant time-step information. Technical indicators such as the Relative Strength Index (RSI), the Moving Average Convergence Divergence (MACD), and Bollinger Bands are also employed to improve feature richness and market dynamics perception. HCLA is compared with cutting-edge deep learning models like standalone LSTM, CNN+LSTM, and multi-layer CNN+LSTM models. The comparison, based on Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE), shows that HCLA has a 12.4% lower MSE, 10.7% lower RMSE, and 11.3% lower MAE than the top-performing deep learning baseline. The close similarity between expected and actual stock prices proves the excellence of the model's accuracy and stability for real-time stock prediction scenarios.
Bhanujyothi et al. (Mon,) studied this question.