<![CDATA[Stock price prediction remains challenging due to the market’s nonlinear, volatile nature, influenced by diverse economic and behavioral factors. Traditional models often suffer from overfitting and limited generalizability. This study addresses these limitations from prior research by other researchers for integrating Long Short-Term Memory (LSTM) with k-Fold Cross-Validation to improve prediction robustness. The proposed framework systematically evaluates model performance across varying market conditions. This methodological contribution enhances forecasting accuracy and stability, offering a more reliable approach to complex financial time series prediction. This study employs LSTM with one to two layers of 64–128 units, trained using Adam and dropout regularization, to capture long-term dependencies in stock price data. The workflow integrates feature selection, Min-Max scaling, and k-Fold Cross-Validation for robust evaluation. Model performance is assessed using RMSE, with reconfiguration applied to address underfitting or overfitting. The proposed model demonstrated substantial performance gains, achieving an average RMSE improvement of approximately 78.40% across all tested stocks compared to prior research. These enhancements are attributed to optimal hyperparameter tuning, consistent use of the Adam optimizer, and the implementation of k-Fold Cross-Validation, which reduced overfitting and provided more stable evaluations. Furthermore, findings revealed that simpler feature sets, such as using only closing prices, can outperform multiple technical indicators when normalization is inadequate, underscoring the importance of robust preprocessing and validation strategies. This study concludes that integrating LSTM with k-Fold Cross-Validation and optimized hyperparameters significantly improves stock price prediction accuracy.]]>
