IoT Traffic Obfuscation: Will it Guarantee the Privacy of Your Smart Home?

Recent research has shown the efficacy of machine learning-based IoT network traffic analysis to infer attributes such as IoT device type, IoT device activity state and even user behaviours in smart home environments. Therefore, various traffic obfuscation techniques have been proposed to reduce the classification performance of these machine learning algorithms. However, most of the proposed traffic obfuscation techniques can only alter traffic originating from the IoT device, with the incoming traffic from the communicating servers largely unaffected. We show that IoT device activity can still be successfully inferred by only using incoming network traffic for analysis. Therefore, this research emphasizes the need for obfuscation techniques, which can alter network traffic in both directions between the IoT devices and their communicating servers.