JBCA: Designing an Adaptative Continuous Authentication Architecture

Abstract

Several continuous authentication methods provide very accurate ways of detecting the impersonation of a user. However, even though they should be access control chain's links, they are commonly documented as algorithms unrelated to any system. In this paper, we propose a continuous authentication architecture to integrate several kinds of continuous authentication engines in a complete framework. The architecture covers different use cases where continuous authentication methods, placed in the user's device or in the cloud, collaborate to determine if the user's identity has changed. We use ensemble learning principles to combine the values generated by each CA model, and compute the trust level adaptively, addressing common errors caused by changes in the environments of continuous authentication systems.

Comparing Machine Learning Classifiers for Continuous Authentication on Mobile Devices by Keystroke Dynamics

“…This paper implemented and compared different ML agent models for CA in mobile environments. Although specific, scalable architectures have been presented [32], to the best of our knowledge, this is the first implementation and testing with specific models for keystroke dynamics using different classifiers on the same dataset. The results suggest that ensemble classifiers (RFC, ETC, and GBC) work better for the problem at hand than instance-based algorithms (k-NN), hyperplane methods (SVM), Bayesian models (naïve Bayes), and decision trees (CART).…”