Many people use smart-home devices, also known as the Internet of Things (IoT), in their daily lives. Most IoT devices come with a companion mobile application that users need to install on their smartphone or tablet to control, configure, and interface with the IoT device. IoT devices send information about their users from their app directly to the IoT manufacturer's cloud; we call this the "app-to-cloud way". In this research, we invent a tool called IoT-app privacy inspector that can automatically infer the following from the IoT network traffic: the packet that reveals user interaction type with the IoT device via its app (e.g., login), the packets that carry sensitive Personal Identifiable Information (PII), the content type of such sensitive information (e.g., user's location). We use Random Forest classifier as a supervised machine learning algorithm to extract features from network traffic. To train and test the three different multi-class classifiers, we collect and label network traffic from different IoT devices via their apps. We obtain the following classification accuracy values for the three aforementioned types of information: 99.4%, 99.8%, and 99.8%. This tool can help IoT users take an active role in protecting their privacy.
Keywords: IoT; IoT privacy inspector; privacy; supervised machine learning.