Applying machine learning in credit risk pricing: Opportunities and challenges for the global financial system

This study examines the evolution of credit risk pricing models in the context of rapid developments in big data and machine learning. The purpose is to evaluate how modern machine learning techniques can improve credit risk pricing compared with traditional approaches. To achieve this, the study develops a comparative analytical framework that contrasts conventional models, such as logistic regression and the Merton structural framework, with modern algorithms, including boosting methods, deep neural networks, and hybrid models. The analysis assesses model performance in terms of predictive accuracy, interpretability, stability, and cross-market adaptability. The findings indicate that machine learning models significantly outperform traditional approaches, with the area under the ROC curve (AUC) reaching up to 0.90 even during periods of economic volatility. The study also incorporates explainability tools, including SHAP and LIME, to clarify the decision mechanisms of complex models and enhance transparency in line with Basel III and IFRS 9 requirements. Cross-regional testing further shows that machine learning models maintain relatively stable performance across different economic environments. Overall, the study concludes that integrating traditional credit risk modeling with modern data-driven techniques can strengthen credit risk management and provide practical implications for financial institutions operating in an increasingly data-intensive global financial system.