Enhancing Phishing Email Detection through Multi Modal Deep Learning

Masireddy, Srihitha (2025) Enhancing Phishing Email Detection through Multi Modal Deep Learning. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

Phishing emails continue to pose significant cybersecurity threats, with sophisticated attacks evading traditional detection methods through careful manipulation of both content and structure. Current approaches typically focus on either semantic analysis or structural patterns in isolation, missing the complementary signals that could enhance detection accuracy. This research presents a novel hybrid architecture that integrates XLNet's advanced language understanding capabilities with Graph Neural Networks' structural pattern recognition, unified through cross-modal contrastive learning.

The proposed system processes emails through parallel pathways, extracting 768- dimensional semantic features via fine-tuned XLNet and 128-dimensional structural features through heterogeneous graph analysis. A contrastive learning framework aligns these representations in a shared 256-dimensional space, enabling effective integration of complementary information. The implementation evaluates four fusion strategies, with attention-based combination achieving optimal performance at 98.71% accuracy.

Comprehensive evaluation against traditional machine learning baselines and ablation studies validates the architecture's effectiveness. While the improvement over XLNet alone is modest at 0.69 percentage points, the research demonstrates successful integration of multi-modal features for phishing detection. The system maintains sub-500ms processing time per email, meeting practical deployment requirements.

This work contributes to cybersecurity by demonstrating how cross-modal learning can unite semantic and structural analysis, providing a foundation for future research in multi-modal phishing detection. The modular architecture and comprehensive evaluation establish benchmarks for hybrid approaches in email security.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Milosavljevic, Vladimir UNSPECIFIED
Subjects:	P Language and Literature > P Philology. Linguistics > Computational linguistics. Natural language processing Z Bibliography. Library Science. Information Resources > ZA Information resources > ZA4150 Computer Network Resources > The Internet > Electronic Mail T Technology > TK Electrical engineering. Electronics. Nuclear engineering > Telecommunications > The Internet > Electronic Mail Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	26 Aug 2026 08:48
Last Modified:	26 Aug 2026 08:48
URI:	https://norma.ncirl.ie/id/eprint/9643

Actions (login required)

View Item