An integrated CSPPC and BiLSTM framework for malicious URL detection.

Area of Science:

• Cybersecurity
• Machine Learning
• Deep Learning

Background:

• Phishing attacks are increasingly diverse, necessitating robust detection methods.
• Existing machine learning and deep learning models for phishing URL detection often have limitations in accuracy.
• Accurate detection of malicious URLs is crucial for overall cybersecurity.

Purpose of the Study:

• To propose CSPPC-BiLSTM, a novel malicious URL detection model.
• To enhance the accuracy and robustness of phishing website detection.
• To leverage attention mechanisms and multi-scale pooling for improved feature extraction.

Main Methods:

• Utilizing Bidirectional Long Short-Term Memory (BiLSTM) for capturing contextual information in URL character sequences.
• Integrating the Convolutional Block Attention Module (CBAM) to highlight key features through channel and spatial attention.
• Employing Spatial Pyramid Pooling (SPP) for multi-scale feature extraction.
• Implementing dropout regularization for enhanced model robustness.

Main Results:

• CSPPC-BiLSTM demonstrated significantly improved detection accuracy compared to the CharBiLSTM baseline.
• The model showed strong generalization and accuracy on both balanced (Grambedding) and imbalanced (Mendeley AK Singh 2020 phish) datasets.
• Ablation experiments validated the crucial contributions of CBAM and SPP modules to performance enhancement.

Conclusions:

• CSPPC-BiLSTM offers a superior approach to malicious URL detection.
• The integration of CBAM and SPP modules is effective in boosting detection performance.
• The proposed model provides a more accurate and robust solution for cybersecurity against phishing attacks.