Touchscreen Biometrics Across Multiple Devices

As the cost of mobile devices decreases, it is becoming increasingly common for users to own more than one. The presence of multiple pieces of mobile technology complicates the question of how to secure them. Utilizing different authentication solutions on different devices may create usability challenges, while using the same authentication technique on more than one device raises the possibility of a compromise of one device affecting the others. Behavioral biometrics, which model the manner in which users interact with their devices, are an appealing option for a single authentication mechanism solution which is capable of working across different devices. Whether or not a user's behavioral features are specific to a particular device is an open question, however. Intuitively, a user's behavior should be independent of what device they are using. In practice, however, this behavior may be impacted by device hardware and software characteristics such as form factor and virtual keyboard layout.

This paper presents an initial investigation into whether or not biometric touchscreen profiles (i.e., trained classification models which can be utilized to authenticate users to their devices) can be applied across more than one mobile device. We conduct a preliminary IRB-approved investigation in which 10 users were asked to perform 3 common tasks on 3 different mobile devices: reading, typing, and playing a game. We then applied the well-known Support Vector Machine (SVM) learning algorithm to touchscreen features collected during each task. The results of this small-scale study indicate that user behavior is consistent for gameplay and reading across dierent types of mobile hardware, but different for typing. This provides preliminary evidence that it is possible to apply behavior-based authentication across multiple devices in some, but not all, contexts.