Enhancing security in IoMT using federated TinyGAN for lightweight and accurate malware detection

Area of Science:

• Cybersecurity
• Machine Learning
• Internet of Things

Background:

• The Internet of Medical Things (IoMT) faces significant malware threats due to numerous connected devices handling sensitive data.
• Inadequate device management creates vulnerabilities, allowing rapid malware propagation with minimal detection.

Purpose of the Study:

• To propose a lightweight and adaptive intrusion detection framework for resource-constrained IoMT devices.
• To enhance malware detection accuracy and efficiency in IoMT environments using federated learning and synthetic data generation.

Main Methods:

• Development of a federated learning (FL) framework integrated with TinyGAN for generating synthetic data to improve malware detection.
• Decentralized, continuous learning approach to adapt to emerging threats without centralized retraining, preserving privacy and reducing computational load.

Main Results:

• The FL-TinyGAN framework achieved high performance metrics: 99.30% precision, 100% recall, and 99.52% F1-score after 20 training rounds.
• Experimental analysis on the IoT-23 dataset showed FL with TinyGAN outperformed traditional models (MLP, FNN/LSTM) in accuracy, convergence, and resource efficiency.

Conclusions:

• The proposed FL-TinyGAN framework offers a scalable, privacy-preserving, and effective solution for securing IoMT environments against malware.
• Demonstrates significant practical advancements in IoT malware detection, outperforming conventional methods in key performance indicators.