Predicting A User's Demographic Identity from Leaked Samples of Health-Tracking Wearables and Understanding Associated Risks
Sudip Vhaduri, Sayanton Vhaduri Dibbo, Chih-You Chen · 2022 IEEE 10th International Conference on Healthcare Informatics (ICHI) · 2022
Recently, market wearables, such as smartwatches with their powerful sensing and computing capabilities, are helping us to real-time monitoring activity, sleep quality, mental health, heart and blood disorders, among many others. Ad-ditionally’ these wearables support various services, including unlocking smartphones or cars, checking emails, and managing financial payments. Since most of these services are based on users' personal data, such as demographics, researchers have been trying to secure those personal data using authentication mechanisms based on biometric, such as heart rate, step count, calorie burn, and blood oxygen saturation values. However, leaking these biometric samples can reveal a user's demographic information, leading to user vulnerability both in cyberspace and physical space. In this work, we demonstrate that biometric samples from single and paired wearables can be utilized to predict a user's two types of attributes or demographic information, i.e., discrete or categorical attribute (ethnicity) and continuous attribute (age) using classifiers and regression models. From our detailed analysis of 40 subjects, we can predict the ethnicity of an unknown user with an average accuracy of up to 0.73 ± 0.06 and the age of an unknown user with an average root mean square error of up to 7.85 + 0.14.