بكارگيري فرايندهاي عصبي مبتني‌بر تبديل‌گر براي پيش‌بيني قيمت در بازارهاي مالي

Financial markets, particularly the global gold market (XAU/USD), pose significant challenges for accurate short-term price forecasting due to their highly volatile nature an‎d complex price patterns. This study focuses on the design an‎d eva‎luation of an innovative model for predicting gold price fluctuations within a 15-minute timeframe. Developed as a novel approach tailored to the characteristics of financial markets, the model draws inspiration from evidencial conditional neural processes (ECNPs) an‎d transformer architectures, integrating deep learning capabilities with uncertainty modeling in a unified framework. By leveraging selec‎ted technical indicators, the model demonstrates a high capacity for identifying an‎d extracting intricate market patterns. A key strength of this approach is its ability to train on relatively limited data, delivering reliable predictions of overall trends using only a small subset of available information. This capability significantly mitigates constraints arising from data scarcity an‎d the complexity of volatile financial markets. Furthermore, the proposed model effectively estimates uncertainties stemming from data structures an‎d limitations in model knowledge, while utilizing transformer encoder mechanisms to accurately capture both short-term an‎d long-term price movement dependencies. Given the volatile nature of financial markets an‎d the challenge of limited data, the model successfully combines the strengths of neural networks an‎d stochastic processes to not only detect complex price patterns but also provide more stable predictions. Experiments conducted under real market conditions demonstrate that the proposed model outperforms approaches based on ensemble learning an‎d regression in terms of prediction accuracy an‎d transaction management reliability. These findings underscore the significance of the research in offering a novel approach to forecasting in volatile financial markets, highlighting the modelʹs potential to enhance trading decision-making an‎d risk management, while paving the way for future applications of neural processes in financial

5

150120