Z2F: Heterogeneous graph-based Android malware detection.

Area of Science:

• Cybersecurity
• Machine Learning
• Mobile Security

Background:

• Android malware poses a significant threat to smart devices and user data.
• Existing detection methods often overlook high-order, hidden information within applications.
• This limitation hinders effective identification of sophisticated malicious behaviors.

Purpose of the Study:

• To propose Z2F, a new framework for detecting Android malware.
• To leverage multidimensional feature extraction and graph neural networks (GNNs).
• To uncover high-order hidden semantic information indicative of malicious activities.

Main Methods:

• Z2F extracts seven types of Android features from application files.
• Features are embedded into a heterogeneous graph for analysis.
• Meta-structures and a multi-layer graph attention mechanism are employed to mine hidden information.

Main Results:

• The study analyzed 14,429 Android applications, extracting over a million features.
• The Z2F framework achieved a remarkable detection accuracy of 99.7%.
• High-order hidden semantic information was effectively mined.

Conclusions:

• Z2F demonstrates superior performance in Android malware detection.
• The framework's ability to mine hidden information is crucial for identifying advanced threats.
• This approach offers a promising direction for enhancing mobile security.