Big Data, Behavioral Biometrics, and the Era of Continuous Authentication - A Review

Abstract

The combination of pervasive sensing, machine learning and large-scale data collection of human behavior has brought a paradigm change in digital identity assurance from one shot authentication to a continuous authentication process. This review aims to provide a synthesis of the peerreviewed literature, systematic surveys, industry telemetry and regulatory developments that span the period from approximately 2015 to 2026 in order to characterize this transition. It explores the big data underpinnings of behavioral biometrics, including the volume, velocity and variety of the interaction signals e.g., keystroke dynamics, touch and swipe gestures, gait, mouse movements, and contextual device usage, and the machine learning architectures that transform these signals into a continuous identity confidence score. The review then examines the documented performance of the paradigm, key technical limitations behavioral drift, variability across devices, adversarial imitation and generative attacks, and lack of standardized benchmarks, and unique privacy and governance issues, as behavioral biometrics can be used to make inferences about health, emotion, and demographics, which are distinct from physiological biometrics. The main claim is that continuous authentication has progressed much further technically than evaluatively and ethically the field boasts impressive equal error rates on curated data, but lacks standardized adversarial benchmarks, longitudinal field evidence, and consent frameworks commensurate with a technology whose very nature is that users can't tell when it's working. The review ends with a research agenda on privacy preserving architectures, disaggregated evaluation and regulatory alignment.