Context-Aware Active Authentication Using Smartphone Accelerometer Measurements

Abstract

While body movement patterns recorded by a smartphone accelerometer are now well understood to be discriminative enough to separate users, little work has been done to address the question of if or how the position in which the phone is held affects user authentication. In this work, we show through a combination of supervised learning methods and statistical tests, that there are certain users for whom exploitation of information of how a phone is held drastically improves classification performance. We propose a two-stage authentication framework that identifies the location of the phone before performing authentication, and show its benefits based on a dataset of 30 users. Our work represents a first step towards bridging the gap between accelerometer-based authentication systems analyzed from the context of a laboratory environment and a real accelerometer-based authentication system in the wild where phone positioning cannot be assumed.