Hybrid Graph Neural Networks With Temporal Attention for Cost-Effective Fraud Detection in Financial Transaction Networks
IEEE Open Journal of the Computer Society · 2026 · 7 · 901-913
DOI: 10.1109/ojcs.2026.3689312 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Open Journal of the Computer Society, 2026. Verfasst von Ofonime Dominic Okon und 4 weiteren Autor:innen. Die Arbeit nennt 51 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The detection of financial fraud faces major obstacles because transactions exist in networked systems while fraudulent methods change quickly and because fraud cases appear in less than 5% of available data.The current methods operate independently on transactions because they do not utilize the connected patterns which appear between linked accounts and time-based sequences. The system proposes Graph Convolutional Neural Networks to extract structural features and Bidirectional Long Short-Term Memory networks with temporal attention to detect sequential patterns and Triplet Loss optimization operates within a multi-task learning system. The framework implements cost-sensitive optimization to manage financial discrepancies which stem from wrong classification results and it uses graph augmentation methods which include edge perturbation and node feature masking and adaptive subgraph sampling to enhance system stability. Our method achieves an F1-score of 0.6052 on the IEEE-CIS dataset which contains 590,540 transactions and 394 features thus outperforming baseline methods by 6.16%. The model achieves its most important benefit by lowering operational expenses for each transaction from $15.84 to $6.78 which results in a 57.2% cost reduction that shows statistical significance through Wilcoxon signed-rank tests ($p < 0.001$). The ablation studies demonstrate that the three learning components in the model work together to achieve better results. The framework operates at inference speeds which stay under 98 milliseconds when running on GPU equipment thus enabling real-time deployment for financial organizations.
Zitieren
APA
Okon, O. D., Stephen, B. U., Asuquo, P., Ekpo, O., & Ofosu, M. A. (2026). Hybrid Graph Neural Networks With Temporal Attention for Cost-Effective Fraud Detection in Financial Transaction Networks. IEEE Open Journal of the Computer Society, 7, 901-913. https://doi.org/10.1109/ojcs.2026.3689312
Vancouver
Okon OD, Stephen BU, Asuquo P, Ekpo O, Ofosu MA. Hybrid Graph Neural Networks With Temporal Attention for Cost-Effective Fraud Detection in Financial Transaction Networks. IEEE Open Journal of the Computer Society. 2026;7:901-913. doi:10.1109/ojcs.2026.3689312
BibTeX
@article{Okon2026,
author = {Okon, Ofonime Dominic and Stephen, Bliss Utibe-Abasi and Asuquo, Philip and Ekpo, Otuekong and Ofosu, Michael Asante},
title = {Hybrid Graph Neural Networks With Temporal Attention for Cost-Effective Fraud Detection in Financial Transaction Networks},
journal = {IEEE Open Journal of the Computer Society},
year = {2026},
volume = {7},
pages = {901--913},
doi = {10.1109/ojcs.2026.3689312}
}
EUVIMED