Machine learning based detection of covert communications under jamming interference.

Area of Science:

• Wireless Communications
• Network Security
• Machine Learning Applications

Background:

• Detecting covert communications in large-scale Internet of Things (IoT) networks is crucial for preventing infrastructure misuse.
• Challenges like fading, noise, jamming, and unpredictable traffic hinder reliable detection by monitoring nodes.
• Existing methods struggle with robustness in interference-limited environments.

Purpose of the Study:

• To develop a robust hybrid analytical-machine learning (ML) framework for detecting covert signals in IoT networks.
• To address challenges posed by jamming interference and Rayleigh fading conditions.
• To provide a scalable and power-agnostic solution for securing wireless networks against covert threats.

Main Methods:

• Derived an analytical energy detection model to establish baseline detection probabilities.
• Utilized Monte Carlo simulations for dataset generation based on the analytical model.
• Extracted signal components as features for supervised training of Decision Tree (DT) and Random Forest (RF) classifiers.

Main Results:

• Both ML models (DT and RF) outperformed the analytical benchmark in detecting covert signals.
• The Random Forest (RF) classifier achieved a 26.8% reduction in total detection error at 1 km.
• The framework demonstrated robustness across varying transmitter and jammer power levels, confirming effectiveness in interference-limited scenarios.

Conclusions:

• The hybrid framework effectively integrates theoretical modeling with data-driven ML for robust covert signal detection.
• Machine learning classifiers, especially Random Forest, offer superior performance compared to traditional analytical methods.
• This approach provides a scalable, power-agnostic solution for enhancing the security of real-world wireless networks against covert threats.