Automated Android Malware Detection Using User Feedback.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• Mobile applications (apps) are widely used, making them targets for malware. App marketplaces employ automated detection, but some malicious apps bypass these systems. Existing detection methods use static or dynamic app features.
• Malware poses risks by accessing sensitive data or blocking user access. Apps evading initial detection can remain available, endangering users. This highlights a gap in current mobile security strategies.

Purpose of the Study:

• To develop a machine learning-based approach for detecting mobile malware that bypasses existing marketplace security.
• To leverage user-provided data, such as quantitative ratings and alert flags, as features for malware classification.

Main Methods:

• Utilized real-world data from an app store, focusing on apps that evaded initial malware detection.
• Trained machine learning classifiers using user-generated quantitative ratings and alert flags as input features.
• Evaluated the accuracy of the trained classifiers in identifying previously undetected malicious applications.

Main Results:

• The machine learning classifiers successfully identified mobile malware that had bypassed standard detection mechanisms.
• The approach demonstrated reasonable accuracy in classifying evaded malware, indicating its effectiveness.
• User feedback data proved valuable for training effective malware detection models.

Conclusions:

• User feedback, including ratings and flags, can be effectively used to train machine learning models for detecting sophisticated mobile malware.
• This method enhances the security of app marketplaces by identifying and classifying malicious apps that evade automated detection.
• The study contributes to maintaining a safer mobile environment for users by improving malware detection capabilities.