Deep learning hybridization for improved malware detection in smart Internet of Things.

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Internet of Things (IoT)

Background:

• The proliferation of AI-enabled IoT devices generates massive big data, creating significant security vulnerabilities and decision-making challenges.
• Existing security measures struggle to cope with the dynamic nature and scale of IoT data, impacting privacy and organizational resources.

Purpose of the Study:

• To introduce a specialized BERT-based Feed Forward Neural Network Framework (BEFNet) tailored for securing AI-enabled IoT environments.
• To address the challenges of dynamic decision-making with continuously growing big data in IoT ecosystems.

Main Methods:

• A novel framework, BEFNet, utilizing a BERT-based Feed Forward Neural Network, was developed for IoT security analysis.
• The framework was optimized using the Spotted Hyena Optimizer (SO) and evaluated on 8 distinct malware datasets.
• Performance was assessed using metrics including accuracy, Matthews Correlation Coefficient, F1-Score, AUC-ROC, and Cohen's Kappa.

Main Results:

• BEFNet achieved exceptional performance metrics across diverse malware datasets.
• Key performance indicators include 97.99% accuracy, 97.96% Matthews Correlation Coefficient, 97% F1-Score, 98.37% AUC-ROC, and 95.89% Cohen's Kappa.
• The Spotted Hyena Optimizer enhanced BEFNet's adaptability to various malware data structures.

Conclusions:

• BEFNet represents a robust defense mechanism for the evolving landscape of IoT security.
• The framework offers an effective solution for dynamic decision-making challenges posed by big data in IoT environments.
• This research highlights the potential of specialized AI frameworks in bolstering cybersecurity for interconnected devices.