Abstract: The development of digital technology has accelerated the transformation of financial online payment methods, bringing convenience to payment but also increasing the hidden dangers of fraudulent transactions. Anti-fraud research is particularly essential to protect users’ property and prevent financial crises. With the advancement of data governance and sharing technology, digital payment transaction data present new characteristics of massive, multi-source and heterogeneous. Integrating data intelligence technology based on big data and artificial intelligence into anti-fraud research has important theoretical research significance. The digital credit payment model formed by the full combination of credit card payment and digital payment has the most mature data accumulation and theoretical basis at present, providing the most ideal data resources and theoretical support for the research of anti-fraud models. Starting from the concept, this paper firstly introduces the definition, research difficulties, and data framework of the digital credit anti-fraud research problem in combination with the actual business scenarios in China. Secondly, based on the modeling strategy, the frontier progress of digital credit transaction anti-fraud research is reviewed from two aspects of data balance and model optimization. This paper focuses on the theoretical basis, applicable scenarios, and latest achievements of various machine learning algorithms and deep learning algorithms in anti-fraud research, and based on the above content, a comprehensive evaluation is made. Finally, combined with the research status and from the perspective of demand, this paper summarizes the three major hotspots including the generalization and interpretability of anti-fraud research, and the sensitivity to new fraudulent transaction models, and concludes with an outlook on future research directions.

Key words: digital credit payment, fraudulent transaction identification, data intelligence, imbalance classification

摘要： 数字技术的发展加速了金融在线支付方式的转变，带来支付便捷的同时却也增加了欺诈交易的隐患，反欺诈研究对保护用户财产、防范金融危机尤为重要。伴随数据治理与共享技术的进步，数字支付交易数据呈现海量、多源、异构的新特点，将基于大数据与人工智能的数据智能技术融入到反欺诈研究中具有重要的理论研究意义。信用卡支付与数字支付充分结合发展形成的数字信用支付模式，拥有当下最成熟的数据积累和理论基础，为反欺诈模型的研究提供了最理想的数据资源与理论支持。从概念出发，首先结合我国实际业务场景，对数字信用反欺诈研究问题的定义、研究难点、数据框架进行介绍；其次基于建模策略，分别从数据均衡和模型优化两方面对数字信用交易反欺诈研究的前沿进展进行综述，重点介绍了各类机器学习算法与深度学习算法在反欺诈研究中的理论基础、适用场景、最新成果，并基于上述内容展开综合评估；最后结合研究现状，从需求的角度切入，对包含反欺诈研究的泛化性、可解释性、面对新型欺诈交易模式敏感性在内的三大研究热点进行总结，并对未来的研究方向进行展望。

关键词: 数字信用支付, 欺诈交易识别, 数据智能, 不均衡分类