Advancing Phishing Email Detection: A Comparative Study of Deep Learning Models.

Area of Science:

• Cybersecurity
• Machine Learning
• Artificial Intelligence

Background:

• Phishing attacks pose significant threats to individuals and organizations.
• Existing email phishing detection methods require improvements in accuracy and reduced false-positive rates.

Purpose of the Study:

• To investigate one-dimensional CNN-based models (1D-CNNPD) for detecting phishing emails.
• To enhance the base 1D-CNNPD model by augmenting it with recurrent layers (LSTM, Bi-LSTM, GRU, Bi-GRU).

Main Methods:

• Developed and experimented with augmented 1D-CNNPD models incorporating recurrent layers.
• Evaluated model performance on two benchmark datasets: Phishing Corpus and Spam Assassin.
• Utilized metrics such as precision, accuracy, F1 score, and recall.

Main Results:

• Augmenting the 1D-CNNPD base model generally improved performance.
• The 1D-CNNPD model combined with Bi-GRU demonstrated superior results.
• An advanced 1D-CNNPD with Leaky ReLU and Bi-GRU achieved 100% precision, 99.68% accuracy, 99.66% F1 score, and 99.32% recall.
• Model depth influences performance, with initial improvements followed by a decline.

Conclusions:

• Augmented 1D-CNNPD models are effective for detecting phishing emails with high accuracy.
• The Bi-GRU augmented model shows significant potential for improving cybersecurity solutions.
• The study highlights the efficacy of deep learning approaches in combating email phishing.