Behavior-Based Probabilistic User Identification Through Passive Sensing

A notion of identity is of vital importance to each individual and the society. Its significance is manifest from two perspectives. On one hand, many systems rely on identity to provide proper services and fulfill their functionality. On the other hand, loss or leakage of identity can cause disastrous consequences. Identity is hence the sought-after to both benign and malicious entities. The representation, acquisition, validation and application of identity have been evolving consistently with the advances in technology. The last decade has witnessed the rapid adoption of mobile electronic devices and the emergence of Internet-of-Things hardware. Their onboard solid-state sensors enable ubiquitous sensing of user, environment and the interaction between them, accumulating ample sensory data that can be leveraged for user identification. In this dissertation, we study behavior-based user identification through passive sensing and its application in new real-world scenarios. We apply statistical modeling methods to the sensory measurements and extract sufficient entropy to establish user identity. A corresponding behavior-based identification framework is specified with necessary components and steps. We contrast the behavior-based approach to existing mechanisms and highlight its advantages. To demonstrate the practical implications of the proposed approach, we propose two application scenarios in the mobile and IoT realms, and conduct experiments. The mobile application investigates user recognition across mobile devices. We leverage the way users interact with mobile apps to track them even when they switch between multiple mobile devices. The IoT application, following the emergence of IoTequipped buildings, studies room-level indoor localization. We model residents’ mobility patterns through occupancy measurements and further learn their locations. Our experimental results demonstrate the effectiveness of identifying users through behavioral patterns under distinct application scenarios.