EUVIMEDEUVIMED entdecken
Crossref

Tri-Branch Deep Representation Learning for Financial Transaction Anomaly Detection Using Behavioral and Deviation-Aware Features

N. Ravi, S. Sathiamoorthy

VFAST Transactions on Software Engineering · 01.08.2026 · 14 (3) · 161-178

DOI: 10.21015/vtse.v14i3.2482 ↗

Kurz gefasst

Fachartikel in der Zeitschrift VFAST Transactions on Software Engineering, online erschienen am 01.08.2026. Verfasst von N. Ravi und 1 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Financial institutions utilize financial transaction anomaly detection systems to identify anomalous financial transactions in the large-scale ecosystems. Financial fraud detection and anti-money laundering (AML) are closely related yet distinct financial crime detection tasks. Financial fraud detection focuses on identifying unauthorized or deceptive transactions, whereas AML concentrates on recognizing transaction patterns related to laundering of illicit funds. Though deep learning boosts representation learning, state of art approaches remains constrained by feature redundancy, imbalance sensitivity, and inadequate anomaly-aware learning mechanisms. For resolving these issues, this article presents a Tri-Branch Deep Learning Framework for Financial Transactions Anomaly Detection (TBDL-FTAD) approach. The presented model incorporates structured feature engineering strategy which combines raw transaction attributes, user-specific behavioral features, and deviation-aware features to improve suspicious transaction detection. In addition, a tri-branch deep learning model is used where context-specific, behavioral, and anomaly-sensitive embeddings are derived to enable enhanced discrimination between legitimate and illicit financial activities. Moreover, a multi-objective loss function is utilized to jointly optimize classification performance, anomaly separation, and sparsity constraints, allowing for more discriminative fraud detection. The experimental result analysis of the TBDL-FTAD technique takes place using two different benchmark datasets, such as PaySim and IBM AMLSIM. The simulation results reported the enhanced outcomes of the TBDL-FTAD technique, gaining an accuracy of 97.98% and 96.08% on both datasets compared to existing methodologies, ensuring its effectiveness as a robust, scalable, and intelligent solution for next-generation AML and financial transaction security analytics. The source code of this study is publicly available at https://github.com/zgenz1537-code/AMLFraudDetection.git

Zitieren

APA

Ravi, N., & Sathiamoorthy, S. (2026). Tri-Branch Deep Representation Learning for Financial Transaction Anomaly Detection Using Behavioral and Deviation-Aware Features. VFAST Transactions on Software Engineering, 14(3), 161-178. https://doi.org/10.21015/vtse.v14i3.2482

Vancouver

Ravi N, Sathiamoorthy S. Tri-Branch Deep Representation Learning for Financial Transaction Anomaly Detection Using Behavioral and Deviation-Aware Features. VFAST Transactions on Software Engineering. 2026;14(3):161-178. doi:10.21015/vtse.v14i3.2482

BibTeX
@article{Ravi2026,
  author = {Ravi, N. and Sathiamoorthy, S.},
  title = {Tri-Branch Deep Representation Learning for Financial Transaction Anomaly Detection Using Behavioral and Deviation-Aware Features},
  journal = {VFAST Transactions on Software Engineering},
  year = {2026},
  volume = {14},
  number = {3},
  pages = {161--178},
  doi = {10.21015/vtse.v14i3.2482}
}
Kostenlos starten →