Credit Risk early warning Research based on Multisource Heterogeneous Data

Accurate early warning of credit risk is crucial for financial stability. Existing models mainly rely on financial indicators, while unstructured data is used less frequently, which limits their predictive performance. This paper constructs multi-source heterogeneous data (MHD) that integrate financial variables, market-attention measures, and corporate-governance attributes, and employs both econometric and machine learning models to systematically examine the gains in credit-risk early warning performance. The SHAP method is further used to quantify feature importance and enhance model interpretability. Results show that, compared with traditional methods, the early warning model enriched with MHD achieves significant improvements in precision, accuracy, F1-score, and AUC. While financial indicators remain the most influential, market attention and corporate-governance factors also play a critical role.