Abstract :

Phishing Websites Remain A Pervasive And Highly Damaging Cybersecurity Threat, Continuously Changing To Bypass Traditional Detection Mechanisms. To Effectively Handle These Threats Without Incurring Massive Computational Overhead, This Paper Presents A Streamlined, Transformer-based Deep Learning Model Implemented Via The TensorFlow Keras Functional API. The Proposed Architecture Focuses Exclusively On Uniform Resource Locator (URL) Token Sequences, Utilizing A Simplified Multi-Head Attention Mechanism Combined With Global Average Pooling To Perform Binary Classification. By Preprocessing URL Strings Into Padded Numerical Sequences And Mapping Them Through A Constrained Embedding Space, The Model Successfully Captures The Syntactic Anomalies Typical Of Malicious Links. Based On An Illustrative Training Phase Across A Representative Dataset, The Architecture Demonstrates Rapid Convergence, Achieving A Validation Accuracy Of Approximately 99.5% Within Just Five Epochs. The Results Underscore The Viability Of Lightweight Self-attention Networks For Real-time Security Applications, Offering A Highly Efficient Alternative To Complex Multi-agent Language Models And Traditional Recurrent Neural Networks. Keywords: Website Phishing, Deep Learning, Transformer-based Model, Cybersecurity, TensorFlow

Published:

18-5-2026

Issue:

Vol. 26 No. 5 (2026)

Page Nos:

1264-1267

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite