Leveraging XAI in Prompt-Based ChatGPT for Financial Decision Support

Neural network models have demonstrated good results in credit evaluation and related fields. However, they are characterized as "black boxes" due to their lack of interpretability. To enhance transparency and credibility, various Explainable Artificial Intelligence (XAI) methods have been proposed. Yet, the raw outputs provided by these XAI methods are still challenging for non-experts to understand. This study presents an innovative approach that applies Large Language Models (LLMs) to credit evaluations. We design contextual prompts that integrate XAI raw outputs into the instructions for LLMs, which then generate domain-specific, and non-expert-friendly explanations in natural language. Experimental results show that the proposed method is capable of generating straightforward explanations of machine learning models for users in different roles.