Online learning classifier based behavioral biometrie authentication
In this paper, we implement a behavioral biometric-based smartphone authentication technique with online training methods. The most obvious difference between online and traditional classification is that online methods allow model training and data collection to be in progress simultaneously. Therefore, our proposed authentication system is superior when dealing with time series data and its model can be updated to adapt the change of user's habit. To verify the feasibility of online training methods used in behavioral biometric authentication, three online algorithms and four traditional classification algorithms are tested with collected dataset. They all achieved an accuracy over 96%. Two additional experiments are designed to test the stability of this authentication system. The results show that the system has the ability to prevent targeted attack, such as shoulder surfing. Furthermore, it keeps high performance that accuracy greater than 95% when user holds smartphone in different scenarios, sitting and walking. Finally, we conclude that online training methods based behavioral biometric smartphone authentication system is very stable with targeted attack and different unlock scenarios.
