Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explainable Transformer-Based Email Phishing Classification with Adversarial Robustness

Sajad U P

TL;DR

This work tackles the rise of AI-generated phishing by developing a robust, explainable detection framework. It combines DistilBERT-based phishing classification with Fast Gradient Method adversarial training and a hybrid explainability pipeline (LIME for feature attribution plus Flan-T5-small for plain-language narratives) to produce accurate, transparent decisions. Key results show improved accuracy ($0.965$) and AUC ($0.985$) over a strong baseline, along with notable robustness to synthetic character-level perturbations. The approach delivers user-friendly explanations that align with common phishing cues, supporting safer decision-making and practical deployment in real-world email security systems.

Abstract

Phishing and related cyber threats are becoming more varied and technologically advanced. Among these, email-based phishing remains the most dominant and persistent threat. These attacks exploit human vulnerabilities to disseminate malware or gain unauthorized access to sensitive information. Deep learning (DL) models, particularly transformer-based models, have significantly enhanced phishing mitigation through their contextual understanding of language. However, some recent threats, specifically Artificial Intelligence (AI)-generated phishing attacks, are reducing the overall system resilience of phishing detectors. In response, adversarial training has shown promise against AI-generated phishing threats. This study presents a hybrid approach that uses DistilBERT, a smaller, faster, and lighter version of the BERT transformer model for email classification. Robustness against text-based adversarial perturbations is reinforced using Fast Gradient Method (FGM) adversarial training. Furthermore, the framework integrates the LIME Explainable AI (XAI) technique to enhance the transparency of the DistilBERT architecture. The framework also uses the Flan-T5-small language model from Hugging Face to generate plain-language security narrative explanations for end-users. This combined approach ensures precise phishing classification while providing easily understandable justifications for the model's decisions.

Explainable Transformer-Based Email Phishing Classification with Adversarial Robustness

TL;DR

This work tackles the rise of AI-generated phishing by developing a robust, explainable detection framework. It combines DistilBERT-based phishing classification with Fast Gradient Method adversarial training and a hybrid explainability pipeline (LIME for feature attribution plus Flan-T5-small for plain-language narratives) to produce accurate, transparent decisions. Key results show improved accuracy () and AUC () over a strong baseline, along with notable robustness to synthetic character-level perturbations. The approach delivers user-friendly explanations that align with common phishing cues, supporting safer decision-making and practical deployment in real-world email security systems.

Abstract

Phishing and related cyber threats are becoming more varied and technologically advanced. Among these, email-based phishing remains the most dominant and persistent threat. These attacks exploit human vulnerabilities to disseminate malware or gain unauthorized access to sensitive information. Deep learning (DL) models, particularly transformer-based models, have significantly enhanced phishing mitigation through their contextual understanding of language. However, some recent threats, specifically Artificial Intelligence (AI)-generated phishing attacks, are reducing the overall system resilience of phishing detectors. In response, adversarial training has shown promise against AI-generated phishing threats. This study presents a hybrid approach that uses DistilBERT, a smaller, faster, and lighter version of the BERT transformer model for email classification. Robustness against text-based adversarial perturbations is reinforced using Fast Gradient Method (FGM) adversarial training. Furthermore, the framework integrates the LIME Explainable AI (XAI) technique to enhance the transparency of the DistilBERT architecture. The framework also uses the Flan-T5-small language model from Hugging Face to generate plain-language security narrative explanations for end-users. This combined approach ensures precise phishing classification while providing easily understandable justifications for the model's decisions.

Paper Structure

This paper contains 32 sections, 2 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Evolution of Phishing Email Detection
  4. Transformer-Based Deep Learning Architectures
  5. Adversarial Robustness and Training
  6. Interpretability in Cybersecurity
  7. Research Gaps and Motivations
  8. Methodology
  9. Data Acquisition and Splitting
  10. Preprocessing Pipeline
  11. Model Architecture
  12. DistilBERT for Phishing Email Classification
  13. FGM-based Adversarial Training
  14. Hybrid Explainability Framework
  15. Model's Decision using LIME
  16. ...and 17 more sections

Figures (2)

  • Figure 1: Common phishing tactics used in email attacks.
  • Figure 2: Overall workflow of the proposed Explainable Transformer-Based Email Phishing Classification with Adversarial Robustness framework