Hybrid IF–GNN–IF Framework for Suspected Fraud Detection in Bank Switching Transactions

Abstract

Fraud detection in financial transactions remains a major challenge in the banking industry, particularly in bank switching systems that process high transaction volumes and involve complex interactions among transaction entities. This study proposes a hybrid Isolation Forest–Graph Neural Network–Isolation Forest (IF–GNN–IF) framework to identify suspected fraudulent transactions in bank switching data. In the first stage, Isolation Forest detects potentially anomalous transactions and produces anomaly scores that are used to construct pseudo-labels. In the second stage, a Graph Neural Network learns relational patterns from a transaction graph by integrating original transaction attributes with the pseudo-labels. In the final stage, the learned graph embeddings are combined with the original transaction features and evaluated using a second Isolation Forest to generate final anomaly scores. The framework was evaluated using 71,873 bank switching transaction records and achieved an accuracy of 0.9668, precision of 0.9200, recall of 0.3681, and an F1-score of 0.5258. The high precision indicates that the framework effectively identifies suspicious transaction patterns. However, the relatively low recall suggests that some suspected fraud cases remain undetected. Therefore, further validation using confirmed fraud labels is necessary to assess the practical effectiveness and generalizability of the proposed framework in real-world banking environments.

Posterior Regimes and Latent Deception: Variational Bayesian Inference in Hidden Markov Models for Sequential Fraud Detection in Financial Transactions
Joseph Uririoghene Obukofe, Anthony O'Hare, Chioma Sandra Dike
arXiv27 Sep 2026PreprintarXiv licenceFull textStatisticsarXiv
A lightweight hybrid CNN–vision transformer framework for image-based Android malware detection
Gopinath Mani, SatheeshKumar Palanisamy, Gnana Sri Manasa Veeramachaneni +3
Journal of Algorithms & Computational Technology23 Oct 2026Peer-reviewedCC BYFull textGeneral mathematicsDOAJ
Transformer-aided guard: A hybrid spatio-temporal framework for confidence-aware intrusion detection in cyber-physical systems
Abeer Aljohani
PLOS One4 Nov 2026Peer-reviewedCC BY 4.0Full textNeuroscience General mathematicsPLOSPMC
Exact random centering and hybrid fluctuation limits for interacting reinforced processes under critical Markov switching
Hugo Cruz-Suárez
arXiv2 Oct 2026PreprintarXiv licenceFull textProbabilityarXiv
Sequential Change Point Detection for PSK Modulation Switching
Gopal Kataria, Davin Sequeira, Arti Yardi
arXiv9 Oct 2026PreprintarXiv licenceFull textGeneral mathematicsarXiv

Cite

BibTeX RIS

BibTeX

@article{muddin2026hybrid,
  title = {{Hybrid IF–GNN–IF Framework for Suspected Fraud Detection in Bank Switching Transactions}},
  author = {Zumam Muddin and Bambang Jokonowo and Afiyati},
  journal = {Journal of Information Systems and Informatics},
  year = {2026},
  month = oct,
  doi = {10.63158/journalisi.v8i5.1799},
  url = {https://doi.org/10.63158/journalisi.v8i5.1799},
  note = {CC BY},
}

RIS

TY  - JOUR
TI  - Hybrid IF–GNN–IF Framework for Suspected Fraud Detection in Bank Switching Transactions
AU  - Zumam Muddin
AU  - Bambang Jokonowo
AU  - Afiyati
PY  - 2026
DA  - 2026/10/24
JO  - Journal of Information Systems and Informatics
DO  - 10.63158/journalisi.v8i5.1799
UR  - https://doi.org/10.63158/journalisi.v8i5.1799
AB  - Fraud detection in financial transactions remains a major challenge in the banking industry, particularly in bank switching systems that process high transaction volumes and involve complex interactions among transaction entities. This study proposes a hybrid Isolation Forest–Graph Neural Network–Isolation Forest (IF–GNN–IF) framework to identify suspected fraudulent transactions in bank switching data. In the first stage, Isolation Forest detects potentially anomalous transactions and produces anomaly scores that are used to construct pseudo-labels. In the second stage, a Graph Neural Network learns relational patterns from a transaction graph by integrating original transaction attributes with the pseudo-labels. In the final stage, the learned graph embeddings are combined with the original transaction features and evaluated using a second Isolation Forest to generate final anomaly scores. The framework was evaluated using 71,873 bank switching transaction records and achieved an accuracy of 0.9668, precision of 0.9200, recall of 0.3681, and an F1-score of 0.5258. The high precision indicates that the framework effectively identifies suspicious transaction patterns. However, the relatively low recall suggests that some suspected fraud cases remain undetected. Therefore, further validation using confirmed fraud labels is necessary to assess the practical effectiveness and generalizability of the proposed framework in real-world banking environments.
SN  - 2656-5935
SN  - 2656-4882
ER  -

Abstract

Related

Cite