ML-Based Device-Agnostic Human Activity Detection with WiFi Sniffer Traffic

Human Activity Detection plays a pivotal role in smoothly managing the health care for the elderly and those with chronic health conditions in smart home environments. Even though there are several technological advancements made in this area, solutions like smartwatches are costly to afford and the solutions that rely on sensors and cameras suffer from privacy concerns. While wireless channel state information-based approaches can address some of these limitations, these approaches either require special hardware to be deployed or modifications at the WiFi access point. In this paper, we propose to detect human activities in a device-agnostic manner by leveraging passively captured passively captured WiFi MAC-layer traffic with the help of a sniffer. In that way, elderly people and those who suffer from chronic health conditions do not need to put any sensors on their body. This approach is not only cost-effective, but it is also easy to deploy without requiring any changes at the WiFi access point or installing special sensors in the environment. We train and test six off-the-shelf machine learning models on 15+ hours worth of WiFi MAC-layer traffic collected in a home environment. We present a proof-of-concept system prototype that employs this approach. We are able to detect six activities - (a) Walking vs Sitting, (b) Sleeping vs Awake, and (c) Using Phone vs Not Using Phone in three different real-world scenarios. Our evaluation reveals that WiFi MAC-layer traffic has special signatures to detect human activities and we are able to achieve 92.49 % detection accuracy in the best case