IoTHound: Environment-agnostic device identification and monitoring

As the Internet of Things (IoT) becomes more ingrained in our daily lives and environments, asset enumeration, characterization, and monitoring become crucial, yet challenging tasks. A vast number of gadgets in the market have a smartphone-based companion-app, making monitoring a variety of devices an overwhelming task for users. We propose IoTHound, an automated method to identify and monitor IoT devices in smart-homes. Our novel prototype leverages capabilities in current commercial off-the-shelf equipment such as routers with multiple antennas that provide insight into the activity of IoT devices in smart homes. We exploit two critical characteristics of IoT networks: device traffic patterns rarely change since devices perform specific tasks, and physical signal properties such as received signal strength indicator (RSSI) are useful since devices can move in closed spaces. IoTHound works without any prior knowledge of the devices. It uses an unsupervised learning method to analyze properties of the network traffic to: (i) identify IoT device types based on extracted network data, and (ii) detect deviations from normal network behavior by monitoring over time. Our evaluation of IoTHound on three distinct datasets comprising Wi-Fi, Bluetooth, Zigbee, and Ethernet devices, indicate that: (i) IoTHound can characterize devices with over 95% accuracy, (ii) IoTHound successfully detects all anomalous behavior in our test scenarios, and (iii) IoTHound can leverage physical characteristics of course device location to enhance its monitoring capabilities.