WristPrint: Characterizing User Re-identification Risks from Wrist-worn Accelerometry Data.

Area of Science:

• Biomedical Engineering
• Computer Science
• Data Privacy

Background:

• Public release of wrist-worn motion sensor data is increasing for health and wellness research.
• This data can be vulnerable to re-identification and sensitive attribute inference attacks.
• Privacy risks for users sharing this data are not fully characterized.

Purpose of the Study:

• To characterize the re-identification risks associated with motion sensor data from wrist-worn devices.
• To guide users and researchers on the privacy implications of sharing such data.
• To quantify re-identification risk based on activity levels.

Main Methods:

• Utilized an open-set deep learning formulation with a novel loss function.
• Trained the model on a dataset of 10 weeks of daily sensor data from 353 users.
• Evaluated re-identification risk in natural user environments.

Main Results:

• Re-identification risk significantly increases with higher user activity intensity.
• On average, a 96% re-identification risk was observed for users sharing a full day of sensor data.
• The study highlights the vulnerability of motion sensor data to privacy breaches.

Conclusions:

• Motion sensor data from wrist-worn devices carries substantial re-identification risks, particularly during high-activity periods.
• Awareness and mitigation strategies are crucial for protecting user privacy in wearable technology research.
• Further research is needed to develop robust privacy-preserving techniques for sensor data.