FinFraudBench: A Heterogeneous Graph Benchmark for Financial Fraud Detection

Yixuan Chen; Hongyu Zhan; Jie Sheng; Weiyu Han; Shuai Chen; Tianyi Zhang; Xiao Tan; Jun Xia

cs.LGAug 15, 2026

FinFraudBench: A Heterogeneous Graph Benchmark for Financial Fraud Detection

Authors: Yixuan Chen, Hongyu Zhan, Jie Sheng, Weiyu Han, Shuai Chen, Tianyi Zhang, Xiao Tan, Jun Xia

Organizations: HKUST-GZ · Jilin University · Ant Group · HKUST

Abstract

The increasing complexity of digital financial systems has reshaped financial fraud detection from isolated transaction classification into relational risk reasoning over interconnected financial entities. This shift has motivated graph-based fraud detection, where models identify fraudulent nodes by exploiting dependencies among customers, cards, merchants, categories, and locations. However, despite rapid progress in graph-based methods, existing public benchmarks remain misaligned with real-world financial systems in two important aspects. First, they often simplify financial ecosystems into homogeneous or single-node-type multi-relational graphs, failing to preserve the multi-entity and multi-relational nature of financial data. Second, they rarely provide large-scale heterogeneous financial graph datasets with realistic operating conditions such as extreme class imbalance and limited label availability, making it difficult to assess the practical effectiveness of current methods. To address these gaps, we present FinFraudBench, a heterogeneous graph benchmark for financial fraud detection. FinFraudBench contains two heterogeneous graph datasets (CreditCard-Fraud and BankTrans-Fraud) with up to 8.99M nodes and 89.23M directed typed edges. Each dataset preserves six financial entity types, fourteen directed edge types, and natural fraud rates that mirror deployment constraints. With these datasets, we establish a standardized evaluation protocol covering both ranking and imbalance-sensitive classification metrics, and evaluate representative baselines. Extensive experiments yield empirical insights into current methods' limitations and suggest promising avenues for future research. FinFraudBench is available at https://anonymous.4open.science/r/FinFraudBench-B002.

Figures & tables

Figure 1: Comparison of three graph formulations for financial fraud detection. Homogeneous graphs use one node type and one relation type; multi-relational graphs keep a single target node type while distinguishing multiple relation types; heterogeneous graphs preserve multiple entity types, multiple relation types, and type-specific attributes.

Figure 2: Overview of the graph construction pipeline. Public transaction-level tables are converted into heterogeneous graphs by mapping transactions and recurring financial objects to typed nodes, deriving semantic relations between observed entities, constructing type-specific attributes, and assigning fraud labels only to transaction nodes.

Figure 3: Transaction split proportions of the constructed fraud graphs.

Dataset	#Trans. nodes	#Total nodes	#Context nodes	#Directed edges	#Fraud trans.	Fraud rate
CreditCard-Fraud	1.85M	1.86M	2.76K	18.53M	9,651	0.5210%
BankTrans-Fraud	8.91M	8.99M	72.14K	89.23M	13,332	0.1495%

Table 1: Graph-scale and label statistics of the constructed datasets.

Dataset	Method	Model family	AUROC	AUPRC	Acc.	Macro-F1	Fraud F1	Fraud Recall
CreditCard-Fraud	MLP	Non-GNN	$0.880{\pm}0.006$	$0.905{\pm}0.005$	$0.754{\pm}0.005$	$0.739{\pm}0.006$	$0.678{\pm}0.009$	$0.517{\pm}0.010$
	LLM-ZS	Non-GNN	$0.659{\pm}0.005$	$0.688{\pm}0.009$	$0.551{\pm}0.003$	$0.452{\pm}0.004$	$0.219{\pm}0.006$	$0.126{\pm}0.004$
	LLM-ICL-4	Non-GNN	$0.807{\pm}0.042$	$0.850{\pm}0.036$	$0.684{\pm}0.012$	$0.650{\pm}0.016$	$0.541{\pm}0.026$	$0.374{\pm}0.025$
	GCN	Homogeneous GNN	$0.820{\pm}0.011$	$0.863{\pm}0.007$	$0.542{\pm}0.003$	$0.420{\pm}0.007$	$0.155{\pm}0.012$	$0.084{\pm}0.007$
	GraphSAGE	Homogeneous GNN	$0.905{\pm}0.006$	$0.928{\pm}0.004$	$0.768{\pm}0.010$	$0.756{\pm}0.012$	$0.700{\pm}0.017$	$0.542{\pm}0.020$
	R-GCN	Multi-relation	$0.887{\pm}0.007$	$0.916{\pm}0.004$	$0.688{\pm}0.007$	$0.654{\pm}0.009$	$0.547{\pm}0.014$	$0.377{\pm}0.013$

Table 2: Overall benchmark results on balanced mini-test subsets. Results are averaged over five mini-test subsets and reported as mean

\pm

standard deviation. The best result for each dataset and metric is highlighted with boldface and gray shading for visual emphasis.

Figure 4: Family-level best results on balanced mini-test subsets. For each metric and model family, bars report the best-performing method within that family in Table 2 ; CC denotes CreditCard-Fraud and BT denotes BankTrans-Fraud.

Figure 5: Family-level mean-score radar across benchmark metrics.

Figure 6: AUPRC–Fraud Recall frontier for trained non-LLM baselines, showing that stronger ranking performance does not always lead to higher fraud recall at the selected decision threshold. Higher values indicate better performance on both axes.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Node type	Representative feature groups
transaction	amount, time fields, location fields, distance or transaction context
customer	profile encodings, location fields, transaction counts and amount statistics
card	billing/context encodings, transaction counts, amount statistics, unique entity counts
merchant	category encoding, transaction counts, amount statistics, unique customer/card/state counts
category	transaction counts, amount statistics, unique merchant/customer/card counts
location	transaction counts, amount statistics, unique entity counts, location/population summaries

Appendix

Table 3: Representative non-label feature groups by node type.

Dataset	Split	#Transactions	#Fraud	#Benign	Fraud rate
CreditCard-Fraud	train_labeled	37.05K	214	36.83K	0.5776%
	train_unlabeled	1.22M	7.08K	1.22M	0.5789%
	validation	37.05K	214	36.83K	0.5776%
	full_test	555.72K	2.15K	553.57K	0.3860%
BankTrans-Fraud	train_labeled	178.30K	268	178.03K	0.1503%
	train_unlabeled	6.78M	10.13K	6.77M	0.1495%

Appendix

Table 4: Detailed transaction split statistics.

Method	AUROC	AUPRC	Acc.	Macro-F1	Fraud F1	Fraud Prec.	Fraud Recall
MLP	$0.8798\pm 0.0057$	$0.9046\pm 0.0052$	$0.7539\pm 0.0050$	$0.7393\pm 0.0059$	$0.6775\pm 0.0087$	$0.9822\pm 0.0020$	$0.5172\pm 0.0103$
GCN	$0.8199\pm 0.0107$	$0.8632\pm 0.0067$	$0.5416\pm 0.0033$	$0.4204\pm 0.0065$	$0.1554\pm 0.0115$	$0.9861\pm 0.0099$	$0.0844\pm 0.0068$
GraphSAGE	$0.9046\pm 0.0059$	$0.9279\pm 0.0044$	$0.7681\pm 0.0102$	$0.7556\pm 0.0118$	$0.7003\pm 0.0167$	$0.9890\pm 0.0044$	$0.5422\pm 0.0195$
R-GCN	$0.8870\pm 0.0072$	$0.9163\pm 0.0044$	$0.6876\pm 0.0066$	$0.6542\pm 0.0088$	$0.5466\pm 0.0137$	$0.9957\pm 0.0030$	$0.3768\pm 0.0128$
HAN	$0.9025\pm 0.0038$	$0.9269\pm 0.0024$	$0.6858\pm 0.0080$	$0.6515\pm 0.0108$	$0.5422\pm 0.0170$	$0.9978\pm 0.0023$	$0.3724\pm 0.0160$
SeHGNN	$0.9026\pm 0.0048$	$0.9279\pm 0.0026$	$0.7575\pm 0.0059$	$0.7426\pm 0.0073$	$0.6807\pm 0.0111$	$0.9959\pm 0.0041$	$0.5172\pm 0.0137$

Appendix

Table 5: Complete balanced mini-test results on CreditCard-Fraud. Values are means and standard deviations across five mini-test subsets.

Method	AUROC	AUPRC	Acc.	Macro-F1	Fraud F1	Fraud Prec.	Fraud Recall
MLP	$0.9384\pm 0.0043$	$0.9456\pm 0.0041$	$0.6718\pm 0.0030$	$0.6329\pm 0.0042$	$0.5133\pm 0.0068$	$0.9926\pm 0.0032$	$0.3462\pm 0.0063$
GCN	$0.9476\pm 0.0032$	$0.9486\pm 0.0030$	$0.7257\pm 0.0052$	$0.7055\pm 0.0065$	$0.6285\pm 0.0099$	$0.9732\pm 0.0051$	$0.4642\pm 0.0111$
GraphSAGE	$0.9263\pm 0.0019$	$0.9302\pm 0.0026$	$0.5460\pm 0.0043$	$0.4286\pm 0.0082$	$0.1696\pm 0.0144$	$0.9915\pm 0.0093$	$0.0928\pm 0.0086$
R-GCN	$0.9615\pm 0.0039$	$0.9717\pm 0.0028$	$0.8542\pm 0.0082$	$0.8510\pm 0.0087$	$0.8292\pm 0.0112$	$1.0000\pm 0.0000$	$0.7084\pm 0.0164$
HAN	$0.9687\pm 0.0028$	$0.9766\pm 0.0021$	$0.8813\pm 0.0043$	$0.8796\pm 0.0044$	$0.8654\pm 0.0055$	$0.9995\pm 0.0012$	$0.7630\pm 0.0088$
SeHGNN	$0.9689\pm 0.0019$	$0.9784\pm 0.0012$	$0.8838\pm 0.0032$	$0.8822\pm 0.0034$	$0.8686\pm 0.0041$	$0.9995\pm 0.0007$	$0.7680\pm 0.0065$

Appendix

Table 6: Complete balanced mini-test results on BankTrans-Fraud. Values are means and standard deviations across five mini-test subsets.

Figure 7: Stability of balanced mini-test evaluation. Heatmaps report the coefficient of variation across five mini-test subsets; lighter cells indicate more stable estimates.

Explore similar work

Graph-Based Fraud Detection with Dual-Path Graph Filtering
Apr 14, 2026Wei He, Wensheng Gan, Philip S. YuFinancial Fraud Detection Graph Representation Learning
School of Intelligent Systems Science and Engineering, Jinan University, Zhuhai 519070, China · Department of Computer Science, University of Illinois Chicago, Chicago 60607, USA
Graph-Based Financial Fraud Detection with Calibrated Risk Scoring and Structural Regularization
May 12, 2026Yunfei Nie, Jiawei Wang, Ruobing Yan +3Graph Neural Networks Graph Representation Learning
Brandeis University, Waltham, USA · University of California, Los Angeles, USA · Georgetown University, Washington, D.C., USA +2
A Multi-Scale Graph Learning Framework with Temporal Consistency Constraints for Financial Fraud Detection in Transaction Networks under Non-Stationary Conditions
Mar 15, 2026Yiming Lei, Qiannan Shen, Junhao SongFinancial Fraud Detection Spatiotemporal GNNs
Georgia Institute of Technology, Atlanta, GA, USA · Boston University, Boston, MA, USA · Imperial College London, London, UK

Apr 14, 2026J/K move · Enter open · S save

Graph-Based Fraud Detection with Dual-Path Graph Filtering

Wei He, Wensheng Gan, Philip S. Yu

School of Intelligent Systems Science and Engineering, Jinan University, Zhuhai 519070, China · Department of Computer Science, University of Illinois Chicago, Chicago 60607, USA

Financial Fraud Detection Graph Representation Learning

Open paper AskPDF arXiv

Fraud detection on graph data can be viewed as a demanding task that requires distinguishing between different types of nodes. Because graph neural networks (GNNs) are naturally suited for processing information encoded in graph form through their message-passing operations, methods based on GNN models have increasingly attracted attention in the fraud detection domain. However, fraud graphs inherently exhibit relation camouflage, high heterophily, and class imbalance, causing most GNNs to underperform in fraud detection tasks. To address these challenges, this paper proposes a Graph-Based Fraud Detection Model with Dual-Path Graph Filtering (DPF-GFD). DPF-GFD first applies a beta wavelet-based operator to the original graph to capture key structural patterns. It then constructs a similarity graph from distance-based node representations and applies an improved low-pass filter. The embeddings from the original and similarity graphs are fused through supervised representation learning to obtain node features, which are finally used by an ensemble tree model to assess the fraud risk of unlabeled nodes. Unlike existing single-graph smoothing approaches, DPF-GFD introduces a frequency-complementary dual-path filtering paradigm tailored for fraud detection, explicitly decoupling structural anomaly modeling and feature similarity modeling. This design enables more discriminative and stable node representations in highly heterophilous and imbalanced fraud graphs. Comprehensive experiments on four real-world financial fraud detection datasets demonstrate the effectiveness of our proposed method.

Abstract

Explore similar work

Graph-Based Fraud Detection with Dual-Path Graph Filtering

Graph-Based Financial Fraud Detection with Calibrated Risk Scoring and Structural Regularization

A Multi-Scale Graph Learning Framework with Temporal Consistency Constraints for Financial Fraud Detection in Transaction Networks under Non-Stationary Conditions