随着大数据时代的到来,信用风险管理在金融领域的重要性日益凸显,信用评分作为其核心工具,面临着海量增长的客户信用数据和个体信用画像动态变迁的挑战。传统的信用评估方法在适应性和灵活性上存在不足,尤其是在处理不平衡数据时。本文提出了基于不平衡数据的AdaFocal-XGBoost集成信用评分模型,旨在提高信用风险预测的准确性和适应性。AdaFocal-XGBoost模型结合了XGBoost的高效计算和AdaFocalLoss的自适应损失调整,特别针对样本不平衡问题进行了优化。通过在UCI数据库中的四个信贷数据集(Australian、German、Japan和Taiwan地区)上的实验,本研究全面评估了AdaFocal-XGBoost模型的性能,并与其他多种信用评分模型进行了对比。结果表明,AdaFocal-XGBoost在AUC、准确率、F1分数、Gmean、KS、精确率和召回率等关键指标上均优于其他模型,特别是在处理严重不平衡数据集时表现出色。本研究不仅验证了集成学习与自适应损失函数结合的有效性,也为信用评分领域提供了新的解决方案,有助于金融机构提高融资效率和管控风险敞口。With the advent of the big data era, credit risk management has become increasingly important in the financial field, and credit scoring, as its core tool, faces the challenge of massive growth of customer credit data and dynamic changes in individual credit profiles. Traditional credit assessment methods are deficient in adaptability and flexibility, especially when dealing with unbalanced data. In this paper, we propose an AdaFocal-XGBoost integrated credit scoring model based on unbalanced data, aiming to improve the accuracy and adaptability of credit risk prediction. The AdaFocal-XGBoost model combines the efficient computation of XGBoost and the adaptive loss adjustment of AdaFocalLoss, which is especially optimized for the sample imbalance problem. Through experiments on four credit datasets (Australian, German, Japan, and Taiwa
