基于动态时序图神经网络的信用卡欺诈检测方法研究

Credit Card Fraud Detection Approach Based on a Dynamic Temporal Graph Neural Network

摘要: 信用卡欺诈是金融科技领域的重大挑战。现有基于静态快照的图神经网络方法难以有效捕捉欺诈行为在连续时间上的动态演化模式。当前研究多将动态交易流离散化为静态图，导致时序信息丢失和检测延迟。本文提出一种新的动态时序图神经网络模型(DTGNN-FD)。该模型采用连续时间动态图范式，通过门控记忆机制和时间感知注意力来持续更新用户表征，从而对流式交易进行实时风险建模。在两个真实金融欺诈数据集上的实验表明，所提模型在F1-Score和Recall等关键指标上优于多种基线模型，表现出识别复杂时序模式的潜力。本研究为实时信用卡欺诈检测提供了一种有效的解决方案。

Abstract: Credit card fraud poses a significant challenge in the field of financial technology. Existing graph neural network methods based on static snapshots struggle to effectively capture the dynamic temporal evolution of fraudulent behaviors. Current research often discretizes dynamic transaction flows into static graphs, resulting in loss of temporal information and detection delays. This paper proposes a novel Dynamic Temporal Graph Neural Network model for fraud detection (DTGNN-FD). The model adopts a continuous-time dynamic graph paradigm, which utilizes a gated memory mechanism and temporal-aware attention to continuously update user representations, enabling real-time risk modeling of streaming transactions. Experiments on two real financial fraud datasets show that the proposed model outperforms multiple baseline models in key metrics such as F1-Score and Recall, demonstrating the potential to identify complex temporal patterns. This study provides an effective solution for real-time credit card fraud detection.