A hybrid stacked bidirectional long short-term memory and Bidirectional Encoder Representations from Transformers model for enhanced phishing email detection

Mohammed, Marwan B.

doi:10.37652/juaps.2025.158955.1369

	A hybrid stacked bidirectional long short-term memory and Bidirectional Encoder Representations from Transformers model for enhanced phishing email detection
Journal of University of Anbar for Pure Science
Volume 20, Issue 1, April 2026, Pages 282-291 PDF (637.93 K)
Document Type: Research Paper
DOI: 10.37652/juaps.2025.158955.1369
Author
Marwan B. Mohammed^*
Department of Computer Science, College of Science, Al-Nahrain University, Baghdad, Iraq
Abstract
Phishing attacks have been among the latest cybersecurity threats as they exploit people’s weaknesses and compromise critical information. Traditional methodologies of phishing detection practices based on machine learning have generally been unsuccessful in offering high performance due to limited generalizability and very high false positive rates. In this paper, a novel hybrid deep learning model is proposed that incorporates Bidirectional Encoding Representation from Transformers (BERT) with Bidirectional Long Short-TermMemory (BLSTM) to increase the detection of suspicious emails within the framework of the proposed model. The model utilizes the contextual understanding of texts from BERT and sequential learning from BLSTM to enhance classification accuracy. Experimental results using an available phishing email dataset show that the proposed approach significantly outperforms the baseline models in F1-score, recall, and precision metrics. Those results highlight the advantages of combining transformerbased embedding and recurrent architectures in phishing detection, providing avenues towards a more robust framework for email security enhancement.
Keywords
BERT; BLSTM; Cybersecurity; Dataset; Phishing email detection

Statistics Article View: 368 PDF Download: 352