<![CDATA[This study explores the application of deep learning techniques for stock price prediction by comparing Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and hybrid CNN–LSTM architectures. We propose a hybrid deep learning model that integrates convolutional layers for local feature extraction with LSTM layers for capturing long-term temporal dependencies in financial time-series data. Historical stock price data of INDF.JK obtained from Yahoo Finance were used to train and evaluate the models. The dataset was preprocessed and transformed into sequential input using a sliding window approach to enable effective time-series learning. Model performance was evaluated using several regression metrics, including Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). Experimental results demonstrate that the proposed hybrid CNN–LSTM model achieves superior prediction performance compared with standalone CNN and LSTM models. The hybrid model records an RMSE of 87.77, MAE of 63.97, and MAPE of 1.02%, while achieving the highest R² score of 0.9759. In comparison, the CNN model produces an RMSE of 96.18 and an R² score of 0.9711, whereas the LSTM model achieves an RMSE of 89.13 with an R² score of 0.9752. These results indicate that the hybrid architecture provides more accurate predictions and better captures the complex patterns in stock price movements. The findings confirm that combining CNN and LSTM architectures enables the model to learn both spatial and temporal representations of financial time-series data. CNN layers effectively identify local patterns within historical price sequences, while LSTM layers capture long-term dependencies that influence future stock prices. Consequently, the hybrid CNN–LSTM framework offers a reliable approach for financial forecasting and has strong potential for practical applications in stock market prediction systems. Future work may incorporate additional technical indicators, sentiment data, or attention-based mechanisms to further enhance prediction accuracy and robustness.]]>
