Effectiveness Evaluation of Big Data Risk Control Models in Consumer Credit: A Comparative Study of Default Prediction Based on XGBoost Algorithm

Abstract

With the rapid development of financial technology and the continuous expansion of the consumer credit market, the accurate identification and prediction of credit risks have become crucial for financial institutions in risk management. This paper focuses on credit risk prediction in the consumer credit market, constructing an XGBoost model based on 168,000 pieces of data from Internet finance company S, and comparing it with models such as logistic regression, SVM, and Gaussian Naive Bayes. The results show that the XGBoost model achieves an accuracy rate of 97% on both the training and validation sets, takes only 0.12 minutes, and has a true positive rate (TPR) of 86.04%, with overall performance significantly superior to logistic regression, SVM, and Gaussian Naive Bayes models. Variable contribution analysis further reveals that "new users", "Sesame Credit", "number of shutdown days", and "Ant Credit Pay Limit" are key factors reflecting overdue behaviors in consumer credit. Based on these core influencing factors, this paper designs corresponding risk control strategies to balance risk control and business efficiency. The research provides a theoretical basis and practical guidance for financial institutions to optimize risk control models, helping to reduce credit risks and promote the sustainable development of consumer credit business.