arxiv: 2604.14235 · v1 · submitted 2026-04-14 · 💻 cs.LG · cs.AI

Graph-Based Fraud Detection with Dual-Path Graph Filtering

Wei He , Wensheng Gan , Philip S. Yu This is my paper

Pith reviewed 2026-05-10 15:35 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords fraud detectiongraph neural networksdual-path filteringheterophilyclass imbalancewavelet transformsimilarity graphensemble learning

0 comments

The pith

Dual-path graph filtering decouples structural anomaly modeling from feature similarity to yield better fraud node representations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Fraud graphs typically feature relation camouflage, high heterophily, and severe class imbalance that cause standard graph neural networks to produce poor node distinctions. The paper presents DPF-GFD, which runs a beta wavelet operator on the original graph to extract structural patterns and separately builds a similarity graph from distance-derived node features before applying an improved low-pass filter. Embeddings from the two paths are fused through supervised learning and passed to an ensemble tree model that scores unlabeled nodes for fraud risk. The design replaces single-graph smoothing with complementary frequency paths so that structural anomalies and feature similarities are modeled independently.

Core claim

DPF-GFD applies a beta wavelet-based operator to the original graph to capture key structural patterns, constructs a similarity graph from distance-based node representations and applies an improved low-pass filter, fuses the resulting embeddings through supervised representation learning, and feeds them to an ensemble tree model; this frequency-complementary dual-path approach explicitly decouples structural anomaly modeling from feature similarity modeling and produces more discriminative and stable node representations on highly heterophilous and imbalanced fraud graphs.

What carries the argument

Frequency-complementary dual-path filtering: one path uses beta wavelets on the input graph for structural anomalies while the second path builds and low-pass filters a similarity graph derived from distance-based node representations.

If this is right

Node representations become more stable and discriminative when structural and similarity signals are processed separately rather than smoothed together.
The fused embeddings improve fraud risk scoring by an ensemble tree model on imbalanced graphs.
Performance gains hold across four real-world financial fraud detection datasets that exhibit heterophily and camouflage.
The method avoids the underperformance typical of single-graph GNN smoothing in relation-camouflaged settings.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Dual-path filtering may transfer to other heterophilous node classification tasks such as spam or anomaly detection in social or transaction networks.
If the distance-based similarity graph proves robust across domains, the approach could lower the need for per-dataset filter tuning in production fraud systems.
Extensions could replace the simple distance construction with learned metrics to further strengthen the feature-similarity path.

Load-bearing premise

Constructing a similarity graph from distance-based node representations and applying an improved low-pass filter will reliably separate fraud patterns without introducing new artifacts or requiring extensive hyperparameter tuning on each dataset.

What would settle it

If single-path ablations using only the wavelet operator or only the low-pass similarity path match or exceed the dual-path model's detection accuracy on the four real-world financial fraud datasets, the claimed advantage of complementary filtering collapses.

Figures

Figures reproduced from arXiv: 2604.14235 by Philip S. Yu, Wei He, Wensheng Gan.

**Figure 1.** Figure 1: Overview of the proposed DPF-GFD framework. The model adopts a dual-path graph filtering design to address relation camouflage, heterophily, and class imbalance in fraud graphs. The structural path enhances multi-frequency structural anomalies on the original graph, while the similarity-based path applies low-pass smoothing on a kNN graph to reconstruct feature consistency. The fused representations are th… view at source ↗

**Figure 2.** Figure 2: Beta wavelet-based kernels in the spectral domain when 𝐶 = 4. of 𝛽 ∗ 𝑝,𝑞 (𝑤) as shown in [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: Results of the ablation study on four datasets. removing each component to assess their contributions to fraudster identification. The three variants tested are: DPF-GFD\O: Removes the adaptive filtering module based on the original graph, retaining only the low-pass filtering module based on the neighbor graph and the ensemble tree classification module. DPF-GFD\K: Removes the low-pass filtering module b… view at source ↗

**Figure 4.** Figure 4: Performance under different values of parameter 𝐶 on four datasets. before classification and use the Uniform Manifold Approximation and Projection (UMAP) algorithm [51] for dimensionality reduction and visualization of high-dimensional features [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗

**Figure 5.** Figure 5: Performance under different values of parameter K on four datasets. (a) GCN (b) PMP (c) BWGNN (d) DPF-GFD Normal (0) Fraud (1) [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗

**Figure 6.** Figure 6: Visualization of the learned node embeddings on the FDCompCN dataset. Wei He et al.: Preprint submitted to Elsevier Page 15 of 17 [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗

read the original abstract

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.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The dual-path beta-wavelet plus similarity-graph filter is a plausible tweak for heterophilous fraud graphs, but the decoupling claim rests on shaky sequencing and the abstract gives no numbers to check it.

read the letter

The paper's main move is to run a beta wavelet on the original graph for structural patterns, build a distance-based similarity graph from the resulting node reps, apply an improved low-pass filter there, fuse the two embeddings, and hand them to an ensemble tree for fraud scoring. That specific pairing is what they present as new for handling relation camouflage and high heterophily in financial graphs. The motivation is solid: standard GNN smoothing often fails when fraud nodes look like their neighbors, so splitting structural anomaly capture from feature similarity makes sense on paper. The pipeline description is straightforward and the fusion step plus tree classifier is a practical choice for imbalanced labels. Credit for targeting a real deployment setting instead of just another benchmark. The soft spot is the central decoupling claim. The abstract sequences the beta wavelet first, then says the similarity graph is built from distance-based node representations. If those representations are the wavelet outputs, the second path is not an independent feature model but a post-processed version of the first; that undercuts the frequency-complementary story and could just produce correlated signals rather than stable separation. No equations, no ablation details, and no quantitative results appear in the abstract, so it is impossible to verify whether the filters actually complement each other or whether the similarity graph adds artifacts. The experiments are asserted on four datasets but without numbers or tuning info the effectiveness part stays untestable. This is incremental adaptation of wavelet and low-pass ideas rather than a fundamental shift. Readers working on GNNs for fraud or other heterophilous imbalanced graphs could pick up the fusion trick or the beta-wavelet choice, but the work needs the full math and results to stand on its own. I would send it to peer review so referees can check the independence of the paths and the actual gains, but the abstract alone does not yet make a strong case.

Referee Report

1 major / 1 minor

Summary. The paper proposes DPF-GFD, a graph-based fraud detection model that applies a beta wavelet operator to the original graph for structural patterns, constructs a similarity graph from distance-based node representations and applies an improved low-pass filter, fuses the resulting embeddings through supervised representation learning, and feeds them to an ensemble tree model for fraud risk assessment on unlabeled nodes. The central claim is that this frequency-complementary dual-path filtering paradigm explicitly decouples structural anomaly modeling from feature similarity modeling, yielding more discriminative and stable node representations than single-graph smoothing approaches on heterophilous and imbalanced fraud graphs. Effectiveness is asserted via experiments on four real-world financial fraud datasets.

Significance. If the dual-path design truly delivers independent complementary signals, the approach could provide a targeted improvement for fraud detection on challenging graphs where standard GNN message passing fails due to heterophily and camouflage. The use of an ensemble tree on top of learned embeddings is a pragmatic choice that aligns with common practice in the domain. However, the lack of any quantitative results, ablation studies, or implementation specifics makes it impossible to assess whether the claimed gains are realized or reproducible.

major comments (1)

[Abstract and method description] Abstract and method description: the abstract states that the beta wavelet operator is applied first to the original graph, after which a similarity graph is constructed 'from distance-based node representations' before the improved low-pass filter. If these representations are the wavelet outputs (as the sequential wording indicates), the second path is not an independent feature-similarity model but a post-processed variant of the first; this dependence directly contradicts the central claim of 'explicitly decoupling structural anomaly modeling and feature similarity modeling' and risks correlated rather than complementary signals in heterophilous graphs.

minor comments (1)

[Abstract] Abstract: states that 'comprehensive experiments on four real-world financial fraud detection datasets demonstrate the effectiveness' but supplies no metrics, baselines, ablation results, or implementation details on filter tuning or graph construction, preventing any assessment of the claimed superiority.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed review and constructive comment on our manuscript. We address the concern about potential dependence between the dual filtering paths below and commit to revisions that strengthen the clarity of our claims without altering the underlying method.

read point-by-point responses

Referee: [Abstract and method description] Abstract and method description: the abstract states that the beta wavelet operator is applied first to the original graph, after which a similarity graph is constructed 'from distance-based node representations' before the improved low-pass filter. If these representations are the wavelet outputs (as the sequential wording indicates), the second path is not an independent feature-similarity model but a post-processed variant of the first; this dependence directly contradicts the central claim of 'explicitly decoupling structural anomaly modeling and feature similarity modeling' and risks correlated rather than complementary signals in heterophilous graphs.

Authors: We appreciate the referee identifying this source of potential misinterpretation in the abstract. The sequential wording was chosen for narrative flow and does not reflect the actual design: the distance-based node representations for constructing the similarity graph are computed directly from the original node feature vectors (via a chosen distance metric such as Euclidean distance), before any wavelet filtering occurs. The beta wavelet operator is applied in a separate, parallel path to the original graph to extract structural patterns. These inputs remain independent, allowing the paths to produce complementary signals as stated in the central claim. We will revise the abstract to replace the sequential phrasing with explicit language indicating that the similarity graph uses the initial features and to note the parallel nature of the two paths. This revision will be included in the next manuscript version. revision: yes

Circularity Check

0 steps flagged

No circularity: dual-path design is an explicit architectural choice, not a reduction to fitted inputs or self-citations

full rationale

The paper describes a sequential pipeline (beta wavelet on original graph, followed by similarity-graph construction from distance-based representations and low-pass filtering) as a deliberate design to achieve frequency complementarity. No equations are provided that would allow any claimed prediction or node representation to reduce algebraically to the inputs by construction. No self-citations are invoked to justify uniqueness or to smuggle in an ansatz; the method is presented as a new proposal whose performance is evaluated empirically on external datasets. The derivation chain therefore consists of independent modeling decisions rather than tautological re-use of fitted quantities.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard assumptions about graph neural networks and wavelet transforms being suitable for fraud graphs; no new entities are postulated and no free parameters are explicitly named in the abstract.

axioms (1)

domain assumption Message-passing GNNs can be improved for heterophilous graphs by applying complementary frequency filters on original and similarity graphs
Invoked when claiming the dual-path design enables more discriminative representations

pith-pipeline@v0.9.0 · 5540 in / 1320 out tokens · 44381 ms · 2026-05-10T15:35:25.669689+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages

[1]

Financialfraud:Aliteraturereview

ArjanReurink. Financialfraud:Aliteraturereview. InContemporary TopicsinFinance:ACollectionofLiteratureSurveys,pages79–115. Wiley, 2016

work page 2016
[2]

Graph neural networks for financial fraud detection: A review.Frontiers of Computer Science, 19(9):199609, 2025

Dawei Cheng, Yao Zou, Sheng Xiang, and Changjun Jiang. Graph neural networks for financial fraud detection: A review.Frontiers of Computer Science, 19(9):199609, 2025

work page 2025
[3]

Financial fraud detection using graph neural networks: A systematic review.Expert Systems with Applications, 240:122156, 2024

Soroor Motie and Bijan Raahemi. Financial fraud detection using graph neural networks: A systematic review.Expert Systems with Applications, 240:122156, 2024

work page 2024
[4]

Bolton and David J

Richard J. Bolton and David J. Hand. Statistical fraud detection: A review.Statistical Science, 17(3):235–255, 2002

work page 2002
[5]

Financial fraud detection through the application of machine learning techniques: a literature review

Ludivia Hernandez Aros, Luisa Ximena Bustamante Molano, Fer- nando Gutierrez-Portela, John Johver Moreno Hernandez, and Mario Samuel Rodríguez Barrero. Financial fraud detection through the application of machine learning techniques: a literature review. Humanities and Social Sciences Communications, 11(1):1–22, 2024

work page 2024
[6]

Internet financial fraud detection based on graph learning

Ranran Li, Zhaowei Liu, Yuanqing Ma, Dong Yang, and Shuaijie Sun. Internet financial fraud detection based on graph learning. IEEE Transactions on Computational Social Systems, 10(3):1394– 1401, 2022

work page 2022
[7]

Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and Philip S. Yu. A comprehensive survey on graph neural networks.IEEE Transactions on Neural Networks and Learning Systems, 32(1):4–24, 2021

work page 2021
[8]

Yingtong Dou, Zhiwei Liu, Li Sun, Yutong Deng, Hao Peng, and Philip S. Yu. Enhancing graph neural network-based fraud detectors against camouflaged fraudsters. InThe 29th ACM International ConferenceonInformationandKnowledgeManagement,pages315– 324, 2020

work page 2020
[9]

ASA- GNN: Adaptive sampling and aggregation-based graph neural net- work for transaction fraud detection.IEEE Transactions on Compu- tational Social Systems, 11(3):3536–3549, 2023

Yue Tian, Guanjun Liu, Jiacun Wang, and Mengchu Zhou. ASA- GNN: Adaptive sampling and aggregation-based graph neural net- work for transaction fraud detection.IEEE Transactions on Compu- tational Social Systems, 11(3):3536–3549, 2023

work page 2023
[10]

Global confidence degree based graph neural network for financial fraud detection.Artificial Intelligence and Autonomous Systems, 2(1), 2025

Jiaxun Liu, Yue Tian, and Guanjun Liu. Global confidence degree based graph neural network for financial fraud detection.Artificial Intelligence and Autonomous Systems, 2(1), 2025

work page 2025
[11]

Graph neural network for fraud detection via context encoding and adaptiveaggregation.ExpertSystemswithApplications,261:125473, 2025

Chaoli Lou, Yueyang Wang, Jianing Li, Yueru Qian, and Xiuhua Li. Graph neural network for fraud detection via context encoding and adaptiveaggregation.ExpertSystemswithApplications,261:125473, 2025

work page 2025
[12]

How powerful are spectral graph neural networks

Xiyuan Wang and Muhan Zhang. How powerful are spectral graph neural networks. InThe 39th International Conference on Machine Learning, pages 23341–23362. PMLR, 2022

work page 2022
[13]

Intelligent financial fraud detection practices in the post-pandemic era.The Innovation, 2(4):100176, 2021

Xiaoqian Zhu, Xiang Ao, Zidi Qin, Yanpeng Chang, Yang Liu, Qing He, and Jianping Li. Intelligent financial fraud detection practices in the post-pandemic era.The Innovation, 2(4):100176, 2021

work page 2021
[14]

Reducing false positives in bank anti-fraud systems based on rule induction in distributed tree- based models.Computers & Security, 120:102786, 2022

Ivan Vorobyev and Anna Krivitskaya. Reducing false positives in bank anti-fraud systems based on rule induction in distributed tree- based models.Computers & Security, 120:102786, 2022

work page 2022
[15]

RIFF:Inducingrulesforfrauddetectionfromdecision trees

Lucas Martins, João Bravo, Ana Sofia Gomes, Carlos Soares, and PedroBizarro. RIFF:Inducingrulesforfrauddetectionfromdecision trees. InTheInternationalJointConferenceonRulesandReasoning, pages 50–58. Springer, 2024

work page 2024
[16]

Enhanc- ing banking fraud detection: Role of machine learning and deep learning methods.Premier Journal of Artificial Intelligence, 1(1), 2025

Muhammad Faraz Manzoor and Muhammad Faran Aslam. Enhanc- ing banking fraud detection: Role of machine learning and deep learning methods.Premier Journal of Artificial Intelligence, 1(1), 2025

work page 2025
[17]

Foe for fraud: Transferable adversarial attacks in creditcardfrauddetection

Jan Lum Fok, Qingwen Zeng, Shiping Chen, Oscar Fawkes, and Huaming Chen. Foe for fraud: Transferable adversarial attacks in creditcardfrauddetection. InIEEEInternationalConferenceonWeb Services, pages 286–292. IEEE, 2025

work page 2025
[18]

Detection of fraudulent financial statements based on naïve bayes classifier

Qingshan Deng. Detection of fraudulent financial statements based on naïve bayes classifier. InThe 5th International Conference on Computer Science & Education, pages 1032–1035. IEEE, 2010

work page 2010
[19]

Şahin and Ekrem Duman

Yusuf G. Şahin and Ekrem Duman. Detecting credit card fraud by decision trees and support vector machines. InThe International MultiConference of Engineers and Computer Scientists, volume 1, Wei Heet al.:Preprint submitted to ElsevierPage 16 of 17 Graph-Based Fraud Detection with Dual-Path Graph Filtering pages 442–447. Newswood Limited, 2011

work page 2011
[20]

Financial fraud detection model: Based on random forest

Chengwei Liu, Yixiang Chan, Syed Hasnain Alam Kazmi, and Hao Fu. Financial fraud detection model: Based on random forest. International Journal of Economics and Finance, 7(7):178, 2015

work page 2015
[21]

Bankfrauddetection usingsupportvectormachine

NanaKwameGyamfiandJamal-DeenAbdulai. Bankfrauddetection usingsupportvectormachine. InIEEE9thAnnualInformationTech- nology, Electronics and Mobile Communication Conference, pages 37–41. IEEE, 2018

work page 2018
[22]

KamtaNathMishraandSubhashChandraPandey.Fraudpredictionin smart societies using logistic regression and k-fold machine learning techniques.Wireless Personal Communications, 119(2):1341–1367, 2021

work page 2021
[23]

End-to-end neural network architecture for fraud scoring in card payments.Pattern Recognition Letters, 105:175–181, 2018

Jon Ander Gómez, Juan Arévalo, Roberto Paredes, and Jordi Nin. End-to-end neural network architecture for fraud scoring in card payments.Pattern Recognition Letters, 105:175–181, 2018

work page 2018
[24]

Credit card fraud detectionusingconvolutionalneuralnetworks

Kang Fu, Dawei Cheng, Yi Tu, and Liqing Zhang. Credit card fraud detectionusingconvolutionalneuralnetworks. InInternationalCon- ference on Neural Information Processing, pages 483–490. Springer, 2016

work page 2016
[25]

Amodelbasedonconvolutionalneuralnetworkforonline transaction fraud detection.Security and Communication Networks, 2018(1):5680264, 2018

ZhaohuiZhang,XinxinZhou,XiaoboZhang,LizhiWang,andPeng- weiWang. Amodelbasedonconvolutionalneuralnetworkforonline transaction fraud detection.Security and Communication Networks, 2018(1):5680264, 2018

work page 2018
[26]

Bernardo Branco, Pedro Abreu, Ana Sofia Gomes, Mariana S. C. Almeida, João Tiago Ascensão, and Pedro Bizarro. Interleaved sequence rnns for fraud detection. InThe 26th ACM SIGKDD InternationalConferenceonKnowledgeDiscoveryandDataMining, pages 3101–3109. ACM, 2020

work page 2020
[27]

Research on financial fraud detection based on deep graph neural network

Ningxin Li, Xuyang Zhang, Lidong Xu, and Jianke Zou. Research on financial fraud detection based on deep graph neural network. In The 4th International Conference on Digital Society and Intelligent Systems, pages 382–385. IEEE, 2024

work page 2024
[28]

Fraud detection based on graph neural networks with self-attention

Min Li, Mengjie Sun, Qianlong Liu, and Yumeng Zhang. Fraud detection based on graph neural networks with self-attention. InThe 2nd International Seminar on Artificial Intelligence, Networking and Information Technology, pages 349–353. IEEE, 2021

work page 2021
[29]

BernNet: Learning arbitrary graph spectral filters via bernstein approxima- tion.AdvancesinNeuralInformationProcessingSystems,34:14239– 14251, 2021

Mingguo He, Zhewei Wei, Hongteng Xu, and Others. BernNet: Learning arbitrary graph spectral filters via bernstein approxima- tion.AdvancesinNeuralInformationProcessingSystems,34:14239– 14251, 2021

work page 2021
[30]

Specformer: Spectral graph neural networks meet transformers

Deyu Bo, Chuan Shi, Lele Wang, and Renjie Liao. Specformer: Spectral graph neural networks meet transformers. InThe Eleventh International Conference on Learning Representations, 2023

work page 2023
[31]

Rethinking graph neural networks for anomaly detection

Jianheng Tang, Jiajin Li, Ziqi Gao, and Jia Li. Rethinking graph neural networks for anomaly detection. InInternational Conference on Machine Learning, pages 21076–21089. PMLR, 2022

work page 2022
[32]

American Mathematical Soc., 1997

Fan RK Chung.Spectral graph theory, volume 92. American Mathematical Soc., 1997

work page 1997
[33]

Beta distri- butions.Continuous Univariate Distributions, 2:210–275, 1994

Norman L Johnson, Samuel Kotz, and N Balakrishnan. Beta distri- butions.Continuous Univariate Distributions, 2:210–275, 1994

work page 1994
[34]

Spamreview detection with graph convolutional networks

AoLi,ZhouQin,RunshiLiu,YiqunYang,andDongLi. Spamreview detection with graph convolutional networks. InThe 28th ACM In- ternationalConferenceonInformationandKnowledgeManagement, pages 2703–2711, 2019

work page 2019
[35]

A uni- versal adaptive algorithm for graph anomaly detection.Information Processing & Management, 62(1):103905, 2025

Yuqi Li, Guosheng Zang, Chunyao Song, and Xiaojie Yuan. A uni- versal adaptive algorithm for graph anomaly detection.Information Processing & Management, 62(1):103905, 2025

work page 2025
[36]

Deep learning, 2016

Ian Goodfellow. Deep learning, 2016

work page 2016
[37]

Gadbench: Revisiting and benchmarking supervised graph anomaly detection

Jianheng Tang, Fengrui Hua, Ziqi Gao, Peilin Zhao, and Jia Li. Gadbench: Revisiting and benchmarking supervised graph anomaly detection. InAdvances in Neural Information Processing Systems, pages 29628–29653, 2023

work page 2023
[38]

XGBoost:Ascalabletreeboosting system

TianqiChenandCarlosGuestrin. XGBoost:Ascalabletreeboosting system. InThe 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 785–794, 2016

work page 2016
[39]

Stone.Classification and Regression Trees

LeoBreiman,JeromeH.Friedman,RichardA.Olshen,andCharlesJ. Stone.Classification and Regression Trees. Wadsworth, 1984

work page 1984
[40]

SplitGNN: Spectral graph neural network for fraud detection against heterophily

Bin Wu, Xinyu Yao, Boyan Zhang, Kuo-Ming Chao, and Yinsheng Li. SplitGNN: Spectral graph neural network for fraud detection against heterophily. InThe 32nd ACM International Conference on Information and Knowledge Management, pages 2737–2746, 2023

work page 2023
[41]

Semi-supervised credit card fraud detection via attribute-driven graph representation

Sheng Xiang, Mingzhi Zhu, Dawei Cheng, Enxia Li, Ruihui Zhao, Yi Ouyang, Ling Chen, and Yefeng Zheng. Semi-supervised credit card fraud detection via attribute-driven graph representation. In The AAAI Conference on Artificial Intelligence, pages 14557–14565, 2023

work page 2023
[42]

Leiserson

MarkWeber,GiacomoDomeniconi,JieChen,DanielKarlI.Weidele, Claudio Bellei, Tom Robinson, and Charles E. Leiserson. Anti- moneylaunderinginbitcoin:Experimentingwithgraphconvolutional networks for financial forensics. InACM SIGKDD Workshop on Anomaly Detection in Finance. ACM, 2019

work page 2019
[43]

DGraph: A large-scale financial dataset for graph anomaly detection

XuanwenHuang,YangYang,YangWang,ChunpingWang,Zhisheng Zhang, Jiarong Xu, Lei Chen, and Michalis Vazirgiannis. DGraph: A large-scale financial dataset for graph anomaly detection. In Advances in Neural Information Processing Systems, pages 22765– 22777, 2022

work page 2022
[44]

Kipf and Max Welling

Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. InInternational Conference on Learning Representations, 2017

work page 2017
[45]

Hamilton, Rex Ying, and Jure Leskovec

William L. Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large graphs. InAdvances in Neural Information Processing Systems, pages 1024–1034, 2017

work page 2017
[46]

Graph attention networks

Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph attention networks. InInternational Conference on Learning Representations, 2018

work page 2018
[47]

Pick and choose: a gnn-based imbalanced learning approach for fraud detection

YangLiu,XiangAo,ZidiQin,JianfengChi,JinghuaFeng,HaoYang, and Qing He. Pick and choose: a gnn-based imbalanced learning approach for fraud detection. InThe Web Conference, pages 3168– 3177, 2021

work page 2021
[48]

Partitioning message passing for graph fraud detection

WeiZhuo,ZeminLiu,BryanHooi,BingshengHe,GuangTan,Rizal Fathony, and Jia Chen. Partitioning message passing for graph fraud detection. InThe Twelfth International Conference on Learning Representations, 2024

work page 2024
[49]

Can abnormality be detected by graph neural networks? InThe 31st International Joint Conference on Artificial Intelligence, pages 1945–1951, 2022

Ziwei Chai, Siqi You, Yang Yang, Shiliang Pu, Jiarong Xu, Haoyang Cai, and Weihao Jiang. Can abnormality be detected by graph neural networks? InThe 31st International Joint Conference on Artificial Intelligence, pages 1945–1951, 2022

work page 1945
[50]

Deep graph library: A graph-centric, highly-performant package for graph neural networks

MinjieWang,DaZheng,ZihaoYe,QuanGan,MufeiLi,XiangSong, Jinjing Zhou, Chao Ma, Lingfan Yu, Yu Gai, Tianjun Xiao, et al. Deep graph library: A graph-centric, highly-performant package for graph neural networks. InInternational Conference on Learning Representations Workshop, 2019

work page 2019
[51]

UMAP: Uniform manifold approximation and projection.Journal of Open Source Software, 3(29):861, 2018

Leland McInnes, John Healy, Nathaniel Saul, and Lukas Großberger. UMAP: Uniform manifold approximation and projection.Journal of Open Source Software, 3(29):861, 2018. Wei Heet al.:Preprint submitted to ElsevierPage 17 of 17

work page 2018