Intelligent malware detection based on graph convolutional network.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Traditional malware detection methods (static, dynamic analysis) are increasingly ineffective against advanced anti-detection techniques.
• Artificial Intelligence (AI) offers improved predictive performance for malware detection.
• Malware diversity poses challenges for feature extraction, hindering AI application.

Purpose of the Study:

• To develop an AI-based malware classifier robust to malware characteristic differences.
• To enhance the effectiveness of AI in malware detection despite feature extraction difficulties.

Main Methods:

• Extracting API call sequences from malware code to generate directed cycle graphs.
• Utilizing Markov chain and Principal Component Analysis (PCA) for graph feature map extraction.
• Designing and implementing a Graph Convolutional Network (GCN) based classifier.

Main Results:

• The proposed GCN-based method demonstrates superior performance in most detection scenarios.
• Achieved the highest accuracy ($$) and outperformed existing methods in False Positive Rate (FPR) and accuracy.
• The model exhibits stability in handling the continuous evolution and growth of malware.

Conclusions:

• The GCN-based approach effectively addresses the challenges of malware diversity and feature extraction.
• This method provides a more accurate, stable, and reliable solution for AI-driven malware detection.
• The findings suggest a promising direction for advancing cybersecurity defenses against sophisticated malware.