Cyber-Threat Detection System Using a Hybrid Approach of Transfer Learning and Multi-Model Image Representation.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• Android applications are vulnerable to malicious network traffic, posing risks to sensitive data and critical sectors like banking.
• Existing detection methods may not fully capture the complex nature of evolving malware threats.

Purpose of the Study:

• To develop an advanced malware detection system for Android by integrating textual and visual network traffic features.
• To enhance the accuracy and robustness of malware identification through a multi-modal approach.

Main Methods:

• Utilizing word2vec-based transfer learning to extract textual features from network traffic.
• Employing a malware-to-image algorithm for visualizing network data and extracting texture features using SIFT and ORB.
• Designing a convolutional neural network (CNN) for deep feature extraction and an ensemble model for final classification.

Main Results:

• The proposed system effectively combines textual and texture features for improved malware detection.
• Performance validated on CIC-AAGM2017 and CICMalDroid 2020 datasets, demonstrating high detection capabilities.
• Explainable AI experiments were conducted to interpret the model's decision-making process.

Conclusions:

• The integrated approach of textual and visual analysis offers a powerful method for Android malware detection.
• This system provides a robust solution against sophisticated network-based threats.
• Further research can explore explainable AI to build trust and transparency in malware detection systems.