An entropy-guided hybrid framework for real-time phishing detection in digital communication systems

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Machine Learning

Background:

• Phishing and social engineering pose significant threats to digital communication security.
• Existing detection methods often struggle with scalability and real-time performance.

Purpose of the Study:

• To develop a hybrid framework, CyberSafe-EG-GWO, for effective and efficient detection of phishing-based social engineering threats.
• To enhance the accuracy and reduce the dimensionality of features for real-time phishing detection.

Main Methods:

• Integration of Entropy-Based Normalization Filtering (EBNF) for token selection.
• Utilizing Grey Wolf Optimization (EG-GWO) for discriminative feature selection.
• Employing SpinalNet for efficient classification of digital communications.

Main Results:

• Achieved high accuracy on UCI Phishing Website (95.8%) and CIC-Phishing Email (94.3%) datasets.
• Reduced feature dimensionality by 40-50%, enhancing scalability for real-time deployment.
• Demonstrated robust performance with average inference latency of 0.214 s per sample on GPU.

Conclusions:

• CyberSafe-EG-GWO offers a reproducible, modular approach to real-time phishing detection.
• The framework balances precision, efficiency, and operational robustness for diverse communication environments.
• Potential for further improvements through adaptive thresholding and lightweight embeddings.