Sustainable Delta-Attention Based Transformer for Efficient Phishing URL Detection
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
In recent years, phishing detection has become an essential aspect of cybersecurity, necessitating models that are accurate, sustainable, and efficient for real-time decision-making. Conventional deep learning models such as Recurrent Neural Networks (RNNs), Gated recurrent unit (GRU), Long Short-Term Memory (LSTM) and Transformers are known for their high computational costs and a vast number of trainable parameters. This characteristic renders them impractical for deployment in lightweight or edge computing environments where resource constraints are a significant concern. To address these challenges, our paper proposed a delta-attention-based Transformer model. This model leverages a top-k sparse attention mechanism, optimized by periodic dense sampling, effectively approximating full attention without the heavy computational burden. The proposed attention mechanism streamlines the complexity typically associated with quadratic attention models while retaining the capacity to capture vital semantic information. As a result, our model not only maintains competitive performance but does so with a considerably reduced number of trainable parameters when compared to prevalent standard models. Evaluations conducted using the UCI Phishing Dataset highlight the model’s efficacy, recording a precision of 94% and an AUC of 0.9873. The specialized design of our model supports sustainable deployment, offering an optimal blend of model complexity, accuracy, and interpretability in the realm of phishing URL detection.
Fields of Study & Subjects
Spam and Phishing Detection
Adversarial Robustness in Machine Learning
Hate Speech and Cyberbullying Detection
Publication details
- Journal / source
- International Conference on Consumer Electronics
- Pages
- 1–6
- Publication date
- 3 February 2026
- Language
- EN