Hybrid Graph Neural Networks for Phishing Email Detection Using Transformer-Based Semantic Feature Extraction
International Journal of Innovations in Science and Technology · 01.10.2026 · 8 (8) · 3144
DOI: 10.33411/ijist/2028 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Innovations in Science and Technology, online erschienen am 01.10.2026. Verfasst von Iqra Munir und 2 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Phishing remains a major cybersecurity threat, requiring detection methods that capture both semantic and structural characteristics of email content. Novelty: This study evaluates DistilBERT-based semantic features with GAT, GraphSAGE, and GIN under a common graph-construction and training protocol and further compares soft voting, hard voting, and stacking, extending the closest single-architecture DistilBERT-GAT work. Material and Method: Each CEAS_08 email is represented as a graph of up to 64 token positions using 768-dimensional DistilBERT embeddings and bidirectional edges between consecutive tokens; the three GNNs are trained and evaluated using a stratified 70/15/15 split, with five-fold stratified cross-validation used as a robustness analysis. Result and Discussion: GAT, GraphSAGE, and GIN achieved 98.89%, 99.42%, and 99.51% test accuracy, respectively. Soft voting produced the best reported test performance at 99.52% accuracy and 99.57% F1-score, while the cross-validation analysis yielded 99.38% ± 0.05% mean accuracy. GAT attention analysis also provided interpretable evidence of token-level weighting. Concluding Remarks: The results indicate that combining complementary GNN aggregation strategies with transformer-derived semantic representations can provide highly accurate phishing-email classification under the tested CEAS_08 protocol.
Zitieren
APA
Munir, I., Naseer, I., & Akhtar, S. (2026). Hybrid Graph Neural Networks for Phishing Email Detection Using Transformer-Based Semantic Feature Extraction. International Journal of Innovations in Science and Technology, 8(8), 3144. https://doi.org/10.33411/ijist/2028
Vancouver
Munir I, Naseer I, Akhtar S. Hybrid Graph Neural Networks for Phishing Email Detection Using Transformer-Based Semantic Feature Extraction. International Journal of Innovations in Science and Technology. 2026;8(8):3144. doi:10.33411/ijist/2028
BibTeX
@article{Munir2026,
author = {Munir, Iqra and Naseer, Iftikhar and Akhtar, Shamim},
title = {Hybrid Graph Neural Networks for Phishing Email Detection Using Transformer-Based Semantic Feature Extraction},
journal = {International Journal of Innovations in Science and Technology},
year = {2026},
volume = {8},
number = {8},
pages = {3144},
doi = {10.33411/ijist/2028}
}
EUVIMED