Research on Exchange Rate Prediction Driven by Multi-Source Heterogeneous Data: Based on Retrieval Augmented Generation and Explainable Machine Learning Models

With the accelerating pace of global economic integration, exchange rate fluctuations have been influenced by various uncertain factors. The information explosion has further complicated the factors influencing exchange rate predictions. To fully utilize multi-source heterogeneous data and further optimize exchange rate prediction performance, this paper constructs an exchange rate prediction method based on Retrieval Augmented Generation (RAG) and Explainable Artificial Intelligence models. During the data processing stage, this paper first uses RAG technology to enhance the situational learning ability of Large Language Models (LLMs). By combining prompts and local sentiment databases, LLMs can perform real-time and accurate sentiment analysis on news headlines. Secondly, transfer learning is used to optimize the Inception(v3) model for sentiment analysis of news images, thereby more efficiently analyzing the sentiment variables contained in texts and images, effectively enriching the sources of multi-source heterogeneous data. Extensive experiments have shown that the multi-source heterogeneous dataset constructed in this paper can effectively improve the accuracy of exchange rate predictions and improve the predictive performance of various models across different time steps.