ZeSAI: AI vigilant malware detection in email security with zero shot-based hybrid network and threat intelligence integration

Venkadeshan Ramalingam (), R. Gopal, Syed Ziaur Rahman and R. Senthil
Additional contact information
Venkadeshan Ramalingam: University of Technology and Applied Sciences
R. Gopal: University of Buraimi
Syed Ziaur Rahman: Majan University College
R. Senthil: Majan University College

Journal of Combinatorial Optimization, 2025, vol. 49, issue 5, No 15, 45 pages

Abstract: Abstract In this ever-evolving world of threats, e-mail security is becoming one of the biggest concerns because attackers are constantly searching for new techniques to bypass the existing security measures. Emails containing phishing, malware and other security threats have become far more common place, which is why there is a need to implement new and more efficient adaptive threat detection frameworks. Typically, email security products are outdated within these emerging threats hence the need to evolve into something more effective and smarter in the detection systems. In this regard, Zero Short learning based Artificial Intelligence (ZeSAI)-model is proposed as a new approach to improve threat identification in the context of email security. Initially, to ensure generalization and robust performance, the model uses three broad sets of input data: augmented data based on Context-Preserving Synthetic Email Generation (CPSEG) method and adversarial data, both generated from six datasets and Threat Intelligence feeds offering real-time updates. The proposed ZeSAI model enhances email threat detection through a structured workflow: eXtreme Language Network (XLNet) first generates bidirectional contextual embeddings from email content, capturing nuanced semantic relationships. The Recurrent GRU Network (RGN) then analyses temporal patterns in the email data, identifying complex relationships and variations over time. These RGN-extracted features are integrated with XLNet-generated semantic embeddings in the Cross-Modal Fusion Layer. Finally, Zero-Shot Learning (ZSL) utilizes these combined semantic descriptions and contextual insights to identify new threats based on their similarities to known threats, enabling robust and adaptive threat detection. The proposed approach yields good accuracy and other performance measures; precision, recall, and F1-score; under fivefold and tenfold cross-validation. An ablation study is also carried out to pinpoint the contribution of each module. Specifically, ZeSAI has accuracy of 98.51% in Business Email Compromise (BEC) threat detection, 96.8% in spam detection, 99.18% in phishing detection, 97.2% in malware attachment detection and 98.58% in detecting insider threats.

Keywords: ZeSAI; CPSEG; Threat intelligence; RGN; XLNet; Zero-shot learning (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
http://link.springer.com/10.1007/s10878-025-01306-2 Abstract (text/html)
Access to the full text of the articles in this series is restricted.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:spr:jcomop:v:49:y:2025:i:5:d:10.1007_s10878-025-01306-2

Ordering information: This journal article can be ordered from
https://www.springer.com/journal/10878

DOI: 10.1007/s10878-025-01306-2

Access Statistics for this article

Journal of Combinatorial Optimization is currently edited by Thai, My T.

More articles in Journal of Combinatorial Optimization from Springer
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().