A hybrid DNN-LSTM model for detecting phishing URLs

Area of Science:

• Computer Science
• Cybersecurity

Background:

• Phishing attacks pose a significant threat by impersonating legitimate websites to steal sensitive user information.
• Existing anti-phishing techniques, including machine learning, are crucial for defense, with deep learning showing state-of-the-art results.
• Previous methods often used character embedding or natural language processing (NLP) features in isolation, limiting detection performance.

Purpose of the Study:

• To propose and evaluate novel hybrid deep learning models for enhanced phishing uniform resource locator (URL) detection.
• To investigate the synergistic benefits of combining character embedding and NLP features for improved phishing identification.
• To advance the accuracy and effectiveness of automated phishing detection systems.

Main Methods:

• Development of hybrid deep learning models integrating Long Short-Term Memory (LSTM) and Deep Neural Network (DNN) architectures.
• Utilization of both character embedding features to capture sequential character patterns and NLP features for semantic understanding.
• Evaluation of model performance on dedicated phishing datasets to assess detection accuracy.

Main Results:

• The proposed hybrid deep learning models demonstrated superior performance compared to existing phishing detection models.
• Combining character embedding and NLP features significantly improved the accuracy of phishing URL detection.
• The models effectively leveraged both low-level character sequences and high-level linguistic structures for robust detection.

Conclusions:

• Hybrid deep learning models integrating character embedding and NLP features offer a powerful approach to combat phishing.
• This combined feature strategy significantly enhances the accuracy and effectiveness of phishing detection systems.
• The findings suggest a promising direction for developing more resilient cybersecurity defenses against evolving phishing threats.