Toward supervised shape-based behavioral authentication on smartphones

Currently, smartphone security has received much more attention as users may use their devices to perform various sensitive tasks. For example, users can utilize mobile banking applications for online shopping, which may store many sensitive data on their devices. Hence there is a need to authenticate users and detect imposters. However, traditional textual passwords are easily compromised and are not convenient for users to remember for a long time due to long-term memory limitation. To complement textual passwords, behavioral authentication is developed by authenticating a user based on the relevant biometric features. In this work, we focus on simple shape-based behavioral authentication that requires users to draw shape(s) for authentication, and investigate how to design such kind of behavioral authentication in practice. We consider two research questions: (1) whether the authentication accuracy varies with different shapes, and (2) how many shapes can be used to achieve good usability. In the evaluation, we perform two user studies with 60 participants and measure some typical supervised learning classifiers. Based on the results, we provide insights on designing a supervised shape-based behavioral authentication system, as compared with similar schemes.