Game-Theoretic Obfuscation of Wi-Fi MAC-Layer Traffic Against IoT Device Fingerprinting Attacks

Area of Science:

• Cybersecurity
• Computer Networks
• Machine Learning

Background:

• Smart home Internet-of-Things (IoT) devices are susceptible to passive traffic fingerprinting using MAC-layer metadata.
• Existing defenses like padding and traffic shaping are often vulnerable to machine learning-based attacks.
• Adversaries can identify devices by analyzing encrypted Wi-Fi frames, packet sizes, and inter-arrival times.

Purpose of the Study:

• To propose and evaluate a game-theoretic framework for Wi-Fi MAC-layer cover-traffic injection defenses.
• To introduce and assess a novel defense mechanism: donor-based mimicry injection.
• To compare the effectiveness of donor mimicry against traditional synthetic traffic methods.

Main Methods:

• Developed a game-theoretic framework to model Wi-Fi MAC-layer cover-traffic injection.
• Introduced donor-based mimicry injection, replicating paired device traffic.
• Compared donor mimicry with fixed-rate, exponential, and uniform synthetic baselines using 198 scenarios and eight classifiers.
• Utilized 10-fold cross-validation for robust performance evaluation.

Main Results:

• Donor mimicry at 100% overhead reduced attacker accuracy to 33.5%, while synthetic methods reached 93.9%.
• Behavioral realism, not just injected volume, proved crucial for defense effectiveness.
• A game-theoretic model yielded a mixed-strategy Nash equilibrium, with a deterministic defense achieving 25.9% balanced accuracy.
• Pairing secrecy is essential for the defense's robustness against pairing-aware attackers.

Conclusions:

• Donor-based mimicry injection offers a significant improvement over synthetic cover traffic for defending IoT devices against MAC-layer fingerprinting.
• The defense operates at the access point, requiring no modifications to IoT devices.
• The study highlights the importance of behavioral realism and pairing secrecy for effective network traffic anonymization.