Volume 120

As financial markets become increasingly complex and volatile, the limitations of traditional statistical models in stock price prediction have become more apparent. This paper proposes a hybrid neural network architecture that integrates convolutional feature extraction and attention-based temporal modeling, aiming to address issues such as noise sensitivity, overfitting, and inadequate integration of multimodal data in existing approaches. Through comparative experiments, the model is shown to be effective in enhancing prediction accuracy and robustness. The results indicate that combining temporal dependency modeling with multimodal data fusion represents a promising direction for future financial forecasting. The paper further explores prospective research directions, including multimodal data processing, cross-market analysis, and the development of real-time systems, thereby providing theoretical support for the application of neural networks in the financial domain.

Show more