Credit Rating Model Based on Deep Learning

Credit ratings are essential for distinguishing borrowers’ risk profiles and guiding lending decisions, yet mismatches between ratings and default risks persist, particularly for high-rated borrowers. This study proposes a Gated Recurrent Unit (GRU) model to address this by modeling sequential credit metrics for accurate rating assignments. Using data from 2,157 small and medium enterprise (SME) loans and 2,044 farmer loans (2018–2022), the GRU processes time-series indicators like payment history and debt ratios to classify borrowers into five rating tiers (AAA to B). Principal component analysis (PCA) reduces input dimensionality, enhancing efficiency. The model achieves 86.7% accuracy, outperforming support vector machines (80.4%) and logistic regression (78.9%). It effectively aligns ratings with credit risk, reducing misclassifications in high-rating tiers. Applications include improved bank risk management and policy support for inclusive finance. Limitations involve data granularity and computational costs. This approach offers a dynamic framework for credit rating, advancing financial risk evaluation in China’s lending markets.