IoT Device Fingerprinting: Machine Learning based Encrypted Traffic Analysis

Abstract

Even in the face of strong encryption, the spectacular Internet of Things (IoT) penetration across sectors such as e-health, energy, transportation, and entertainment is expanding the attack surface, which can seriously harm users' privacy. We demonstrate in this paper that an attacker is able to disclose sensitive information about the IoT device, such as its type, by identifying specific patterns in IoT traffic. To perform the fingerprint attack, we train machine-learning algorithms based on selected features extracted from the encrypted IoT traffic. Extensive simulations involving the baseline approach show that we achieve not only a significant mean accuracy improvement of 18.5% and but also a speedup of 18.39 times for finding the best estimators. Obtained results should spur the attention of policymakers and IoT vendors to secure the IoT devices they bring to market.