You are AIRing too Much: Assessing the Privacy of Users in Crowdsourcing Environmental Data

With the availability of inexpensive sensors, the attractiveness of participatory sensing has increased tremendously in the last decade. However, when sensing is performed with devices owned by individuals, it raises several privacy issues with respect to the data producers, and hence reduces the incentive to contribute to the services. In this paper, we evaluate the extent to which a malicious server in a crowdsourcing air quality monitoring service can track the locations of users that contribute to the service. The participants periodically send information, such as temperature, relative humidity, carbon monoxide, and luminosity of their surrounding, using an off-the-shelf sensor connected to their mobile phones. The participants also send their coarse-grain location (i.e., disclosing the ID of the cell tower to which their mobile is coupled) along with the air quality data. We evaluate the precision with which the attacker can track the participants using only air quality data and location of the cell tower. We perform a thorough analysis of the privacy attack and show that it can accurately discover the destination of the users with a precision of more than 85% (up to 97%), if at least five consecutive samples are provided by the participants. We also discovered that the precision drops when the environmental sensors are affected by outside conditions (e.g., exposition to direct sunlight) but remains significant (54.5% for 20 consecutive samples).