Continuous Authentication Using BioAura (CABA) 2.0

Abstract

As the Internet of Things-enabled devices become prevalent, more and more sensors are being devised to monitor humans' daily lives for various applications. Wearable technologies are capitalizing on the innovations in sensor design and the large amount of information collected by wearable medical sensors (WMS) to monitor the health and behavior of users. These sensors and information are rich data sources and can also be leveraged to perform continuous authentication. In most systems, currently, authentication is performed at the point of initial login to determine if the user has the correct credentials to gain access to the system. However, this mechanism suffers from many security flaws; attackers can gain access if the system is left unlocked and unattended. Continuous authentication aims to continuously verify the identity of the user and ensure that he/she is the same person who was originally authenticated. We focus on developing a prototype of a continuous authentication system using BioAura, multiple biological signals collected in realtime from currently existing wearable medical sensors. The sensors used in the prototype collect data passively, noninvasively, and continuously. This work demonstrates that an ensemble of sensors can be used to collect multiple physiological signals and apply robust machine learning models to continuously verify the identity of the user with high accuracy. We collect data from an user study of 30 participants and design the system balancing the tradeoffs between usability and security, ensuring that it is extensible for any authentication application. Finally, we evaluate the authentication accuracy and discuss the threat model of the continuous authentication system. CABA achieves 89\% accuracy on average with ensemble learning methods like AdaBoost, outperforming currently existing wearable devices with only one sensor.