DL-RL: Dual-Layer Graph Neural Networks with Curriculum Reinforcement Learning for Financial Fraud Detection
Keywords:
fraud detection, graph neural networks, reinforcement learning, curriculum learning, transaction graph, PPO, attentionAbstract
Financial fraud imposes hundreds of billions of dollars in annual losses on the global economy and continuously evolves to outpace static detection systems. This paper proposes DL-RL, a Dual-Layer Graph Neural Network framework with Curriculum Reinforcement Learning for real-time financial fraud detection. The lower GNN layer (a Graph Attention Network with 4 heads and 128 hidden units) captures local transaction-neighborhood embeddings via attention-weighted aggregation, while the upper layer (GraphSAGE, 64 units) propagates inter-account risk signals across the global transaction graph. A PPO-based curriculum scheduler progressively exposes the agent to increasingly complex fraudulent subgraphs, growing difficulty from κmin = 0.1 to 1.0 over 500 warm-up episodes. Experiments on three public benchmarks—IEEE-CIS, Elliptic Bitcoin, and PaySim—yield macro F1 scores of 0.931, 0.947, and 0.963 respectively, improving over the strongest baseline by up to 5.2 percentage points while reducing false positive rates by up to 38%. Ablation studies confirm the independent contribution of each architectural component. A proof-of-concept implementation is available at https://github.com/anonymous/DL-RL-fraud (to be released upon acceptance).
References
[1] Association of Certified Fraud Examiners, “Report to the Nations: 2024 Global Study on Occupational Fraud and Abuse,” ACFE, Austin, TX, USA, Tech. Rep., 2024.
[2] T. Albalawi and S. Dardouri, “Enhancing credit card fraud detection using traditional and deep learning models with class imbalance mitigation,” Front. Artif. Intell., vol. 8, p. 1643292, 2025.
[3] S. N. Far et al., “A dual-layer GNN with economic penalty mechanisms for blockchain fraud detection,” Expert Syst. Appl., vol. 286, 2025.
[4] M. Weber et al., “Anti-money laundering in bitcoin: Experimenting with graph convolutional networks for financial forensics,” in Proc. KDD Workshop Anomaly Detection Finance, 2019.
[5] Z. Liu et al., “Mitigating message imbalance in fraud detection with dual-view graph representation learning,” arXiv preprint arXiv:2507.06469, 2025.
[6] H. Zhao et al., “A GNN-based fraud detector with dual resistance to graph disassortativity and imbalance,” Inf. Sci., vol. 671, 2024.
[7] M. Kaur et al., “Leveraging reinforcement learning for real-time fraud detection in financial transactions,” ResearchGate, 2025.
[8] Y. Dong et al., “Dynamic fraud detection: Integrating reinforcement learning into graph neural networks,” arXiv preprint arXiv:2409.09892, 2024.
[9] Y. Cui et al., “FraudGNN-RL: A graph neural network with reinforcement learning for adaptive financial fraud detection,” IEEE Open J. Comput. Soc., 2025.
[10] R. R. Devi, J. E. Raja, and Y. B. Chin, “Reinforcement learning with graph neural network (RL-GNN) fusion for real-time financial fraud detection,” Sci. Rep., vol. 15, p. 42953, 2025.
[11] A. Bhansali et al., “ACLEGR-TADD: Adaptive continual learning for financial fraud detection under extreme class imbalance,” OpenReview, 2025.
[12] W. Hamilton, Z. Ying, and J. Leskovec, “Inductive representation learning on large graphs,” in Proc. NeurIPS, vol. 30, 2017.
[13] P. Velickovic et al., “Graph attention networks,” in Proc. ICLR, 2018.
[14] S. Islam et al., “Detecting fraudulent transactions for different patterns in financial networks using layer weighted GCN,” Hum.-Centric Intell. Syst., vol. 5, pp. 181–195, 2025.
[15] J. Schulman et al., “Proximal policy optimization algorithms,” arXiv preprint arXiv:1707.06347, 2017.
[16] Kaggle / Vesta Corporation, “IEEE-CIS fraud detection dataset,” 2019. [Online]. Available: https://www.kaggle.com/c/ieee-fraud-detection
[17] M. Weber et al., “Elliptic bitcoin dataset,” 2019. [Online]. Available: https://www.kaggle.com/datasets/ellipticco/elliptic-data-set
[18] E. A. Lopez-Rojas and S. Axelsson, “PaySim: A financial mobile money simulator for fraud detection,” in Proc. 28th European Modeling and Simulation Symp., 2016, pp. 249–255.
[19] T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” in Proc. 22nd ACM SIGKDD, 2016, pp. 785–794.
[20] Y. Dou et al., “Enhancing GNN-based fraud detectors against camouflaged fraudsters,” in Proc. 29th ACM CIKM, 2020, pp. 315–324.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Journal of Applied Mathematics, Informatics and Digital Arts
This work is licensed under a Creative Commons Attribution 4.0 International License.
All articles published in this journal are distributed under the terms of the Creative Commons Attribution 4.0 International License (CC BY 4.0).
Under this license, users are free to:
- Share — copy and redistribute the material in any medium or format
- Adapt — remix, transform, and build upon the material for any purpose, including commercial use
Under the following conditions:
- Attribution — Appropriate credit must be given to the original authors and source, a link to the license must be provided, and any changes made must be indicated. Attribution must not suggest endorsement by the authors or the journal.
Author Rights
Authors retain full copyright of their work and grant the journal the right of first publication. After publication, authors may:
- Archive the published version in institutional repositories
- Share and reuse their work with proper attribution
Open Access Statement
This journal follows a Diamond Open Access model, meaning:
- No publication fees (APCs) are charged to authors
- No subscription fees are charged to readers
- All content is freely accessible immediately upon publication
