Can individuals be identified using bio-signal indicators of subjects measured from wearable sensors?

Abstract

In recent years, wearable sensors have been used to non-invasively obtain a variety of biometric information such as heart rate, pulse wave, acceleration, etc. What is the degree of personal identifiability of biometric information obtained from such wearable sensors? In Japan, the revised Personal Information Protection Law, which was fully enforced in 2017, clarified that personal identification codes such as DNA, facial features, and iris are included in personal information. And in recent years, consideration of human rights and privacy for technologies that monitor and track people has been called for. In this study, 14 parameters were extracted from wearable sensor data of 8 subjects over a 4-day period, and their personal identifiability was verified using machine learning. The results suggest that subjects whose biometric values are close to each other may be misidentified, and that subjects with large intra-individual variation in biometric information are significantly less discriminable.