بكارگيري مدل‌هاي بصري براي پيش بيني در بازارهاي مالي

Time series forecasting in financial markets has always been one of the most challenging problems in the fields of artificial intelligence an‎d computational finance, due to the non-stationary nature of the data, high volatility, significant noise, an‎d susceptibility to external factors. Classical statistical methods an‎d many conventional machine learning models face limitations in achieving high accuracy an‎d stable performance, primarily because of their inability to simultaneously capture short-term an‎d long-term dependencies, as well as their ineffective utilization of the visual information inherent in market data. On the other han‎d, market participants in practice heavily rely on charts an‎d visual patterns in financial data, while most forecasting models do not directly incorporate this critical aspect into the learning process. In this research, a deep learning framework based on vision models is proposed for forecasting financial market time series, enabling the extraction of complex price patterns. The proposed architecture leverages convolutional neural networks an‎d modern learning mechanisms to identify local features an‎d recurring market patterns, while simultaneously modeling long-term temporal dependencies effectively through a Transformer structure empowered by the attention mechanism. This approach aims to bridge the gap between human visual analysis an‎d automated forecasting models, providing a scalable an‎d generalizable framework applicable to various financial markets. Empirical eva‎luation results of the proposed model on real financial market data demonstrate significant improvements in forecasting accuracy metrics an‎d trading performance compared to traditional statistical models an‎d common deep learning architectures, such as recurrent neural networks. Additionally, sensitivity analysis of the parameters indicates satisfactory robustness of the model against variations in hyperparameters an‎d market conditions. The findings of this study suggest that vision-based deep learning models can serve as an effective tool for enhancing forecasting accuracy an‎d supporting decision-making in financial markets, while laying a solid foundation for future research advancements.

6

162857