IoT malware detection architecture using a novel channel boosted and squeezed CNN.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• The Internet of Things (IoT) integrates smart devices, creating significant security vulnerabilities.
• Cybercriminals exploit these connected devices using malware, necessitating robust detection methods.
• Malware detection is crucial for safeguarding IoT ecosystems against cyber threats.

Purpose of the Study:

• To propose a novel Convolutional Neural Network (CNN)-based architecture, iMDA, for enhanced IoT malware detection.
• To address the security blind spots in IoT networks caused by interconnected devices.
• To develop a modular and efficient system for identifying and mitigating malware in IoT environments.

Main Methods:

• Developed the iMDA architecture, a modular CNN incorporating edge exploration, multi-path dilated convolutions, and channel squeezing/boosting.
• Implemented split-transform-merge (STM) blocks for learning local structural variations in malware.
• Utilized multi-path dilated convolutions to recognize global malware patterns and channel manipulation for feature diversity.

Main Results:

• The iMDA architecture achieved high performance on a benchmark IoT dataset.
• Key performance metrics include accuracy (97.93%), F1-Score (0.9394), precision (0.9864), and AUC-ROC (0.9938).
• Outperformed several existing state-of-the-art CNN architectures in malware detection.

Conclusions:

• The proposed iMDA demonstrates strong malware detection capabilities for IoT devices.
• The architecture's modular design and feature learning schemes contribute to its effectiveness.
• Future extensions may include composite detection for Android-based malware and IoT Elf files.