With the rapid growth of e-commerce and digital finance, credit cards have become indispensable for consumer transactions but also increase the risks of repayment fraud and defaults, posing challenges to financial institutions, consumer trust, and payment system stability. Traditional risk control methods often struggle to capture complex behavioral patterns in large-scale, high-dimensional transaction data, necessitating more adaptive detection approaches. This study proposes an Adaptive Risk Control Model by comparing Random Forest (RF) and Neural Networks (NN) for credit card default detection, using a dataset of over 200,000 records from 2000 globally active U.S. credit card customers across three major banks between 2005 and 2020, encompassing demographic attributes, credit data, billing statements, and delinquency histories. Results show that the RF model achieved 0.95 accuracy with strong robustness and interpretability, while the NN model achieved 0.93 accuracy, effectively modeling complex feature interactions; both maintained balanced recall and F1-scores despite class imbalance. These findings demonstrate the effectiveness of machine learning in adaptive credit risk detection and provide a scalable, real-time framework for enhancing financial institutions’ risk management systems.
Li et al. (Fri,) studied this question.