Real-Time Email Phishing Detection Using a Custom DistilBERT Model

Damatie, Edafe Maxwell, Eleyan, Amna ORCID: https://orcid.org/0000-0002-2025-3027 and Bejaoui, Tarek (2024) Real-Time Email Phishing Detection Using a Custom DistilBERT Model. In: 2024 International Symposium on Networks, Computers and Communications (ISNCC), 22 October 2024 - 25 October 2024, Washington DC, USA.

Preview

Accepted Version Available under License In Copyright. Download (537kB) | Preview

Abstract

This paper presents a real-time email phishing detection system that utilizes a custom DistilBERT model. The custom DistilBERT architecture incorporates dynamic threshold adjustment and an enhanced classifier head, optimized for analyzing email content. With detection response times of under two seconds, the system delivers real-time protection. Experimental results demonstrate 99.29% accuracy in controlled tests and 95.45% in real-world tests, surpassing current state-of-the-art methods. The system maintains high performance at low false positive rates, essential for practical deployment. An adaptive daily retraining mechanism ensures continued effectiveness against evolving phishing tactics. This research advances email security and offers insights into the use of transformer-based models in real-time cybersecurity applications. By addressing the limitations of current systems through distilled transformer models, this work significantly strengthens organizational cybersecurity against advanced phishing threats.