Graph Based Fraud Detection Dataset for Financial Transaction Networks

This dataset accompanies the research study on graph-based fraud detection using Graph Neural Networks (GNNs). It provides a structured representation of financial transaction activity as a graph, where entities and transactions are represented as nodes and relationships between entities are represented as edges. The dataset is intended to support reproducible research in financial fraud detection, graph machine learning, anomaly detection, and Graph Neural Network-based classification. It contains the data structures and experimental outputs required to reproduce the graph construction and evaluate fraud detection models. The dataset includes transaction-level and graph-level information, processed node and edge representations, fraud labels, and experimental results from GraphSAGE and conventional machine-learning baseline models including Random Forest and XGBoost. To protect privacy and confidentiality, the dataset contains anonymized and/or synthetic representations of transaction activity and does not contain personally identifiable information. The accompanying research project implements graph-based fraud detection using GraphSAGE and compares its performance with conventional machine-learning approaches. The dataset is intended for academic research, benchmarking, methodological comparison, and reproducibility of the associated study.本数据集与一项基于图神经网络（Graph Neural Networks, GNNs）开展图式欺诈检测的研究工作配套发布。该数据集将金融交易活动以结构化图结构形式表征：实体与交易以节点形式呈现，实体间的交互关系以边形式呈现。 本数据集旨在为金融欺诈检测、图机器学习、异常检测以及基于图神经网络的分类任务领域的可复现研究提供支撑。其包含了复现图构建流程与评估欺诈检测模型所需的数据结构与实验输出结果。 本数据集涵盖交易级与图级信息、经过预处理的节点与边表征数据、欺诈标签，以及基于GraphSAGE与传统机器学习基准模型（包括随机森林与XGBoost）生成的实验结果。 为保护隐私与数据机密性，本数据集采用匿名化或合成生成的交易活动表征形式，不包含任何个人可识别信息。 配套研究工作采用GraphSAGE实现图式欺诈检测，并将其性能与传统机器学习方法进行对比。本数据集可用于相关研究的学术研究、基准测试、方法学对比以及结果复现。