Credit Risk Prediction on Imbalanced Data Using Machine Learning

The rapid advancements in internet and big data technologies, coupled with the growing demand for credit services, have introduced significant challenges to traditional credit risk assessment models. To overcome the limitations of conventional approaches in managing data complexity and nonlinear relationships, this study presents an XGBoost-based credit risk prediction model. Leveraging individual and corporate credit data-such as age, gender, marital status, and loan amount-alongside optimized objective functions and regularization strategies, the proposed model achieves notable improvements in predictive performance. Experimental evaluations reveal that XGBoost consistently surpasses traditional models, including decision trees, random forests, and support vector machines, across metrics such as precision, recall, AUC, and accuracy. Its superior performance is particularly evident in scenarios involving imbalanced datasets. These findings highlight XGBoost's capability to effectively process complex financial data and its potential to serve as a robust tool for credit risk management.