Explore the Use of Prompt-Based LLM for Credit Risk Classification

Credit risk assessment plays an important role in financial services by estimating the chance of a borrower defaulting. Recently, although the Large Language Models (LLMs) have demonstrated superior performance in various tasks, especially in natural language processing, their effectiveness in credit risk evaluation remains unknown. Therefore, this study explores the use of prompt-based LLMs for credit risk classification using the “Give Me Some Credit” dataset. The performance of LLM is compared with traditional models, including XGBoost, Support Vector Machine (SVM), Random Forest, and Multi-Layer Perceptron (MLP). The results show that the LLM does not outperform these traditional models in prediction accuracy. However, the LLM offers clear reasoning that can help support decisions. Furthermore, SHAP value analysis highlights the most important features affecting model predictions. Adversarial training also shows that the LLM and XGBoost have similar robustness. These findings suggest that LLMs can be used alongside traditional models to improve transparency and support financial decision-making.