Enhancing Stock Forecasting Accuracy Through LSTM-Based Recurrent Models

Stock market forecasting has long been regarded as one of the most challenging yet critical tasks in financial analytics, as it directly supports decision-making in investment strategies, risk mitigation, and economic planning. Accurate forecasting not only assists individual and institutional investors in optimizing their portfolios but also contributes to the stability and predictability of broader economic systems. Recent developments in deep learning, particularly in the fields of Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM)-based Recurrent Neural Networks (RNNs), have created new opportunities for handling complex financial time-series data. These architectures enable the discovery of non-linear dependencies and hidden structures within large datasets that traditional machine learning approaches often fail to capture. However, despite these advances, a systematic comparative analysis of CNN and LSTM performance across multiple stock indices remains limited, thereby restricting model selection strategies within financial research and practice. In this study, we conduct an extensive evaluation of CNN and LSTM-RNN models by applying them to diverse stock indices, including the Dow Jones Industrial Average (DJI), NASDAQ, New York Stock Exchange (NYSE), Russell 2000, and S&P 500. The experiments reveal that LSTM-based models consistently outperform CNN architectures, achieving predictive accuracies of up to 90 %. The findings of this research highlight the strategic advantage of adopting LSTM-based frameworks for stock market prediction, offering investors, policymakers, and developers reliable tools to enhance predictive performance. Ultimately, the study provides a robust, data-driven comparison between CNN and LSTM approaches, delivering practical guidance for designing future forecasting systems and contributing to the advancement of financial decision-support technologies.