Improving phishing email detection performance through deep learning with adaptive optimization

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Machine Learning

Background:

• Phishing email attacks are increasingly sophisticated, challenging traditional detection methods.
• Distinguishing between legitimate and malicious emails requires advanced techniques due to realistic impersonation.
• Existing cybersecurity measures face limitations against evolving phishing tactics.

Purpose of the Study:

• To propose a novel hybrid deep learning architecture for enhanced phishing email detection.
• To leverage Bidirectional Encoder Representations from Transformers (BERT) for contextual understanding.
• To optimize the model using the Mountain Gazelle Optimizer (MGO) for improved performance.

Main Methods:

• A hybrid architecture combining BERT embeddings, Convolutional Neural Network (CNN), and Gated Recurrent Unit (GRU).
• Incorporation of multi-head attention to refine focus on critical email features.
• Hyperparameter optimization using the Mountain Gazelle Optimizer (MGO) on a Kaggle phishing email dataset.

Main Results:

• Achieved high classification accuracy (96.8%), precision (97.2%), recall (95.4%), and F1 score (96.3%).
• Demonstrated a 2.5% reduction in false positives compared to state-of-the-art methods.
• Validated the model's effectiveness against sophisticated phishing threats.

Conclusions:

• The proposed hybrid deep learning model significantly enhances phishing email detection capabilities.
• Transformer-based embeddings combined with advanced neural networks and optimization are effective against phishing.
• The MGO-optimized architecture offers a robust solution for improving email security.