A Comprehensive Credit Risk Assessment Model: From Risk Identification to Credit Scoring

This paper proposes a comprehensive credit risk assessment framework that integrates feature selection, credit scoring, and credit rating classification to address persistent challenges in the financial industry. In high-dimensional data environments with imbalanced sample distributions, traditional approaches struggle to accurately identify default risk while maintaining model interpretability. We employ Lasso regression for feature selection, Synthetic Minority Over-sampling Technique (SMOTE) combined with random under-sampling for data balancing, and ensemble learning through optimized Random Forest and gradient boosting for credit scoring. Our experimental results on German and Australian credit datasets demonstrate the model's superior performance, achieving accuracy exceeding 92% and AUC reaching 0.946 for credit classification. The framework effectively reduces dimensionality from 24 to 12 key indicators while maintaining predictive power, improves minority class recognition through balanced sampling, and establishes a stable five-level credit rating system using hierarchical clustering and Gaussian Mixture Models. Comparative analysis with deep learning approaches, including neural networks and attention mechanisms, demonstrates competitive performance of our framework while maintaining higher interpretability. This integrated approach provides finan-cial institutions with enhanced risk identification accuracy and transparent decision-making processes adaptable to changing economic conditions.