Aug 2026· Journal of King Saud University: Computer and Information Sciences· Vol 38· 0 citations· 38 references
TL;DR
HND-GFD is proposed, a novel fraud detection framework integrating high-order hypergraph modeling and dynamic neighborhood aggregation that consistently outperforms state-of-the-art baselines and adaptively aggregates neighborhood information from benign and fraudulent perspectives.
Abstract
Graph-based fraud detection, which identifies fraudulent and benign entities on graph-structured data, has shown strong potential in combating sophisticated fraud and attracted growing research attention. However, existing methods face two critical bottlenecks. First, increasingly complex fraud camouflage: fraudsters conceal collusive behaviors via multi-hop connections and deliberately link to benign nodes, preventing traditional models from capturing high-order patterns and causing feature homogenization of fraud nodes. Second, severe class imbalance: fraud nodes account for a tiny proportion of the graph, and weak fraud signals are easily overwhelmed by massive benign node information. To address these challenges, we propose HDN-GFD, a novel fraud detection framework integrating high-order hypergraph modeling and dynamic neighborhood aggregation. Specifically, we design a dual-dimensional hypergraph construction mechanism that upgrades pairwise connections to multi-node collaborative associations along structural and feature dimensions to capture high-order collusive relationships. We then develop an anomaly probability-guided dynamic aggregation strategy, which estimates node anomaly scores via node-subgraph feature consistency and adaptively aggregates neighborhood information from benign and fraudulent perspectives. This design decouples camouflage-induced confounding signals and amplifies minority fraud features, mitigating the adverse impact of class imbalance. Extensive experiments on four real-world datasets demonstrate that HDN-GFD consistently outperforms state-of-the-art baselines, verifying the effectiveness and superiority of our method.
Graph topology and model architecture are routinely co-designed in GNN-based fraud detection, making it impossible to attribute performance gains to either component. We address this by fixing the training loop, features, and evaluation protocol while independently varying the graph construction strategy and GNN architecture. Three strategies are evaluated: multi-relation temporal, hybrid structural similarity, and intra-group, each evaluated across three GNN architectures (GATv2, GCN, and GraphSAGE). A feature-identical MLP with no graph structure serves as an empirical anchor. On the Sparkov dataset, all three GNN strategies exceed the MLP by 5.0–9.7 F1 points, confirming that topology contributes genuine discriminative value when per-cardholder histories are dense. On the IBM dataset, the intra-group strategy collapses 6.9 F1 points below the MLP, a consequence of near-empty neighbourhoods (mean degree 1.1) following down-sampling, while multi-relation retains ranking advantages in AUC (Area Under the Curve; 0.985) and Average Precision (0.908) despite marginal F1 parity. Across both datasets, multi-relation temporal construction is the highest-performing strategy, achieving F1 0.908 and AUC 0.994 on Sparkov and F1 0.831 on IBM, though this ranking is not entirely architecture-independent: GraphSAGE narrowly inverts the multi-relation/intra-group order on Sparkov. These results suggest that graph construction quality, rather than mere graph presence, matters more for GNN performance than connectivity alone under the conditions evaluated here.
Roya Amiri, Sardar F. Jaf· Big Data and Cognitive Compu...· 0 citations
Financial fraud now spreads through coordinated accounts whose risk is visible mainly in how they connect rather than in any single record. Tree based classifiers that score transactions in isolation miss this relational signal. This paper reviews and consolidates how graph neural networks (GNNs) cast fraud and anomaly detection as node and edge classification over transaction graphs, where message passing propagates evidence among accounts, devices, and merchants. We describe the main architectures used in practice, namely graph convolutional networks, GraphSAGE, and graph attention networks, and explain how each aggregates neighbourhood information. Two problems dominate the fraud setting: extreme class imbalance, since genuine fraud is rare, and camouflage, where fraudsters attach to honest neighbours to dilute their signature. We summarise label aware and neighbour selecting designs such as CARE-GNN and PC-GNN that address both. Heterogeneous and temporal graph variants are discussed for settings with several node types and evolving relations, alongside sampling methods that keep training tractable on large graphs. Using illustrative experiments patterned on public benchmarks (Elliptic, Yelp and Amazon review fraud, and IEEE-CIS), we report AUC, AUPRC, F1, and recall at k, and show that imbalance aware GNNs improve minority recall over a gradient boosted baseline. We close with notes on explainability and the engineering needed for production scoring.
Raji N· Eduschool International Jour...· 0 citations
Transaction fraud detection on digital payment platforms poses three inter-twined challenges: sparse supervision, severe class imbalance, and complex relational dependencies among accounts and transaction events. This paper studies fraud detection on a heterogeneous transaction graph constructed from the public PaySim simulator. To this end, we propose ProtoHGC, a framework that integrates heterogeneous graph contrastive pre-training with prototype-regularized classification. ProtoHGC employs a multi-scale en-coder that aggregates first-order neighborhood features and second-order meta-path-based context, coupled with node-level and subgraph-level con-trastive objectives to enhance representation quality under limited supervi-sion. Building on the learned embeddings, class-specific prototypes for normal and fraudulent behaviors enforce intra-class compactness and inter-class separation. Experiments on PaySim demonstrate that ProtoHGC con-sistently outperforms GCN, GAT, GraphSAGE, and MLP baselines across all tested hidden dimensions in terms of AUC-ROC, PR-AUC, accuracy, re-call, and F1-score. We explicitly note that the current evaluation is limited to a single synthetic dataset; few-shot adaptation, cross-domain generaliza-tion, and multimedia-specific fraud scenarios remain unvalidated.
Qi Hu· Poster Volume 0008 The 2026...· 0 citations
: Social network platforms have become primary channels for information dissemination, yet they are increasingly exploited by anomalous users such as bots, fake accounts, and coordinated disinformation spreaders. These malicious actors manipulate public opinion, spread misinformation and undermine platform integrity, posing severe threats to the security of the online ecosystem. Accurate detection of such users is challenging because they often organize into sophisticated high-order connection patterns that extend beyond local neighborhoods. Existing methods address this by either injecting predefined motifs as handcrafted features, which lack flexibility to discover unknown patterns, or employing higher-order Graph neural networks (GNNs) at prohibitive costs. Crucially, neither method treats structural information as learnable knowledge that can be automatically acquired from data and explicitly represented. To bridge this gap, we propose SparseGNN, a structural-knowledge-enhanced framework for anomalous user detection. It regards atomic subgraph patterns as fundamental, learnable units of structural knowledge. This framework is concatenated with original node features and fed into any standard GNN, without modifying the backbone architecture. Experiments on real-world datasets demonstrate that SparseGNN improves the accuracy and F1-score of standard GNNs for anomalous users detection without requiring predefined patterns, while maintaining linear complexity. Because the learned atomic patterns capture global high-order topology, the resulting structural knowledge representation is inherently less sensitive to localized edge perturbations, incidentally conferring improved stability under adversarial structural attacks.
Zehan Li, Yingyi Li, Zhiwei Tang et al.· Computers, Materials & C...· 0 citations
The complexity and volume of transactional data has expanded due to the rapid growth of digital financial services, which has opened the door to new types of sophisticated fraud. The ever-changing nature of fraud trends and the complexity of entity interactions make rule-based or transactional fraud detection systems inadequate. Using relational structure learning and transactional behaviour profiling, FraudGraph-X is a multimodal graph neural framework for behavioural fraud detection. In the financial ecosystem model offered by the framework, users, accounts, devices, and merchants are nodes in a graph. Graph neural networks analyse transactional histories, patterns of activity across time, and ambient metadata to identify group fraud tendencies and high-order dependencies. Unlike conventional methods, FraudGraph-X is able to learn from structural links and behavioural indicators, allowing it to detect complex and organised fraud. Based on thorough experimental testing on real-world and benchmark financial datasets, the system outperforms non-graph deep learning approaches, conventional machine learning, and the detection accuracy of changing fraud tactics. It also has lower false positive rates. Using multimodal graph learning, FraudGraph-X can discover intricate fraud patterns; it is a smart and scalable solution for fraud detection systems of the future.
N. Yatoo, M. Jishnu, Josiah John et al.· ITM Web of Conferences· 0 citations
Financial fraud detection in transaction networks is challenging due to evolving attack strategies, complex relational structures, and extreme class imbalance. We propose a hybrid deep learning model that fuses Graph Convolutional Neural Networks (GCNNs) with bidirectional LSTMs enhanced by temporal attention, enabling joint modeling of structural dependencies and sequential transaction patterns. To improve class discrimination, the framework incorporates Triplet Loss, enhancing prior contrastive approaches, which enhances embedding separability under highly imbalanced conditions. Furthermore, we introduce graph augmentation strategies, including edge perturbation, node feature masking, and adaptive subgraph sampling, to increase robustness against noise and incomplete networks. Experiments on the IEEE-CIS and synthetic datasets with varied patterns demonstrate that the proposed model achieves up to 6.2% improvement in F1-score with a modest precision-recall trade-off favoring high-recall scenarios and 12% higher recall compared to strong baselines. Ablation studies confirm the complementary roles of the graph, temporal, and metric learning components. These consistent improvements, demonstrate that incorporating Triplet Loss within graph–temporal modeling provides a principled and effective approach to imbalanced fraud detection, establishing our model as a scalable and robust solution for fraud detection in financial transaction networks.
Ofonime Dominic Okon, Imo Enang, B. Stephen et al.· E3S Web of Conferences· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
