PermQRDroid: Android malware detection with novel attention layered mini-ResNet architecture over effective...

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• Android OS dominates the global smart device market, with increasing usage driven by open-source and free applications.
• Application installation from unofficial sources poses significant privacy and security risks to users.
• Effective detection of malicious Android applications is crucial for user protection.

Purpose of the Study:

• To develop a novel deep learning model for detecting potentially harmful Android applications.
• To utilize Android application permission information for security threat identification.
• To create a robust and efficient method for Android app security analysis.

Main Methods:

• An attention-layered mini-ResNet model was proposed for image-based Android application analysis.
• QR code-like images were generated using the 100 most effective Android application permissions identified via chi-square technique.
• Residual and attention layers were incorporated to enhance feature extraction and focus on critical permission information.

Main Results:

• The proposed model achieved high accuracy rates (up to 100%) on diverse datasets including Androzoo, Drebin, Genome, and Google Play Store.
• Cross-validation demonstrated consistent performance, with accuracy values ranging from 94% to 98.62%.
• The model outperformed classical machine learning and existing deep learning approaches in classifying Android applications based on permissions.

Conclusions:

• The attention-layered mini-ResNet model offers a highly accurate and efficient solution for Android application security.
• The image-based approach using permission data provides a novel and effective method for threat detection.
• The study highlights the potential of deep learning in enhancing mobile security against evolving threats.