Learning Device Usage in Context: A Continuous and Hierarchical Smartphone Authentication Scheme
Popular smartphone authentication schemes, such as PIN-based or biometrics-based authentication methods, require only an initial login at the start of a usage session to authorize the user to use all the apps on the phone during the entire session. Those schemes fail to provide continuous protection of the smartphone after the initial login. They also fail to meet the hierarchy of security requirements for different apps under different contexts. In this study, we propose a continuous and hierarchical authentication scheme. We believe that a user's app-usage patterns depend on his location context. As such, our scheme relies on app-usage patterns in different location context to continuously establish the log probability density (LPD) of the authenticity of the current user. Based on different LPD thresholds corresponding to different security requirements, the current user either has a LPD higher than the threshold, which grants him continuous access to the phone or the app, or he has a LPD lower than the threshold, which locks him out of the phone or the app immediately. We test our scheme on 4,600 subjects from the Device Analyzer Dataset. We found that our scheme could correctly identify the authenticity of the majority of the subjects. However, app-usage patterns with or without location context yielded similar performances, indicating that user contexts did not contribute further information to establish user behavioral patterns. Based on our scheme, we propose a hypothetical Android app which would provide continuous and hierarchical authentication for the smartphone users.