MVDroid: an android malicious VPN detector using neural networks.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• Virtual Private Networks (VPNs) are widely used for online privacy, but many pose significant security risks.
• Malicious VPNs can steal data, spread malware, and compromise user privacy despite appearing legitimate.
• Android users face particular risks due to the prevalence of untrustworthy VPN applications.

Purpose of the Study:

• To develop an optimized deep learning model for detecting malicious Android VPNs.
• To create a novel dataset of both malicious and benign Android VPN applications for training and evaluation.
• To enhance the security of Android users by identifying and flagging unsafe VPNs.

Main Methods:

• An optimized deep learning neural network was designed and implemented.
• The model was trained and evaluated using a newly curated dataset of Android VPNs.
• App permissions were utilized as key features for identifying malicious VPN behavior.

Main Results:

• The proposed deep learning classifier achieved high accuracy in identifying malicious VPNs.
• The model demonstrated superior performance compared to standard classifiers across key metrics like accuracy, precision, and recall.
• Experimental results validate the effectiveness of the deep learning approach in detecting unsafe VPNs.

Conclusions:

• The developed deep learning model offers a reliable solution for identifying malicious Android VPNs.
• This research contributes to improving the security landscape for mobile users seeking privacy through VPNs.
• The findings underscore the importance of robust security measures in VPN applications.