A Hybrid CNN-BiLSTM-Attention Framework for Privacy-Preserving Fraud Detection on Encrypted Data Using the CKKS Homomorphic Encryption Scheme
International Journal on Advanced Science, Engineering and Information Technology · 22.10.2026 · 16 (5) · 1687-1694
DOI: 10.18517/ijaseit.16.5.14591 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal on Advanced Science, Engineering and Information Technology, online erschienen am 22.10.2026. Verfasst von Wasan Alaa Hussain und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
As machine learning workloads enter the public cloud, there is a structural conflict between scalable analytics and the confidentiality of data being analyzed. A problematic example of this tension is financial fraud detection – the records of transactions include personally identifiable information that the institutions are legally and contractually bound not to reveal, in plain text, to third-party compute providers. In this paper, we propose, implement, and test empirically a pipeline for detecting fraud that integrates a hybrid deep learning classifier with a fully homomorphic encryption scheme, namely Cheon-Kim-Kim-Song (CKKS). In 200 encrypted test transactions, the encrypted pipeline replicates plaintext probabilities with a mean absolute error of 3.29 × 10⁻⁷, and a maximum error of 3.78 × 10⁻⁷, and generates no disagreements in the labels relative to plaintext inferences, demonstrating that the cryptographic computation preserves the classifier output well below any reasonable decision-boundary tolerance. The framework is intended as a practical reference design for institutions requiring to outsource the compute for fraud-screening while preserving mathematical privacy of all transactions being sent to the cloud.
Zitieren
APA
Hussain, W. A., Ahmed, I. S., Kashmoola, M. A., & Anad, M. Y. (2026). A Hybrid CNN-BiLSTM-Attention Framework for Privacy-Preserving Fraud Detection on Encrypted Data Using the CKKS Homomorphic Encryption Scheme. International Journal on Advanced Science, Engineering and Information Technology, 16(5), 1687-1694. https://doi.org/10.18517/ijaseit.16.5.14591
Vancouver
Hussain WA, Ahmed IS, Kashmoola MA, Anad MY. A Hybrid CNN-BiLSTM-Attention Framework for Privacy-Preserving Fraud Detection on Encrypted Data Using the CKKS Homomorphic Encryption Scheme. International Journal on Advanced Science, Engineering and Information Technology. 2026;16(5):1687-1694. doi:10.18517/ijaseit.16.5.14591
BibTeX
@article{Hussain2026,
author = {Hussain, Wasan Alaa and Ahmed, Israa Shihab and Kashmoola, Mohammed A. and Anad, Maan Younis},
title = {A Hybrid CNN-BiLSTM-Attention Framework for Privacy-Preserving Fraud Detection on Encrypted Data Using the CKKS Homomorphic Encryption Scheme},
journal = {International Journal on Advanced Science, Engineering and Information Technology},
year = {2026},
volume = {16},
number = {5},
pages = {1687--1694},
doi = {10.18517/ijaseit.16.5.14591}
}
EUVIMED