Privacy Disclosure from Wearable Devices

Abstract

In recent years, wearable devices have seen an explosive growth of popularity and a rapid enhancement of functionalities. Current off-the-shelf wearable devices offer pack sensors such as pedometer, gyroscope, accelerometer, altimeter, compass, GPS, and heart rate monitor. These sensors work together to quietly monitor various aspects of a user's daily life, enabling a wide spectrum of health- and social-related applications. Nevertheless, the data collected by such sensors, even in their aggregated form, may cause significant privacy concerns if shared with third-party applications and/or a user's social connections (as many wearable platforms now support). This paper studies a novel problem of the potential inference of sensitive user behavior from seemingly insensitive sensor outputs. Specifically, we examine whether it is possible to infer the behavioral sequence of a user such as moving from one place to another, visiting a coffee shop, grocery shopping, etc., based on the outputs of pedometer sensors (aggregated over certain time intervals, e.g., 1 minute). We demonstrate through real-world experiments that it is often possible to infer such behavior with a high success probability, raising privacy concerns on the sharing of such information as currently supported by various wearable devices.