BERT ensemble based MBR framework for android malware detection.

Area of Science:

• Cybersecurity
• Machine Learning
• Internet of Things (IoT)

Background:

• Android malware poses significant risks to IoT devices.
• Existing methods for Android malware detection face challenges in complex IoT environments.

Purpose of the Study:

• To develop a novel framework for Android malware detection (AMD) in recommender systems-based IoT.
• To enhance system security and user privacy against evolving Android malware threats.

Main Methods:

• A BERT Ensemble (MBR) model combined with MobileNetV2 was developed.
• A threat analysis technique using a subset of 100 Android app permissions and a refined feature set was employed.
• Ensemble methods were applied to static data for malware detection.

Main Results:

• The MBR model achieved 98% accuracy, 96% precision, 98% recall, and 97% F1-score.
• The framework demonstrated superior performance compared to MCADS, DroidRL, CNN, FAGnet, GAN, and FEDriod.
• A low log loss of 0.058 was recorded, indicating high model reliability.

Conclusions:

• The MBR framework offers a reliable and innovative solution for Android malware detection in IoT.
• The study addresses critical user privacy and system security concerns in the face of increasing Android malware risks.
• This research presents a novel strategy for malware detection using ensemble methods on static data.