Increasing Accuracy of Hand-Motion Based Continuous Authentication Systems

Most authentication systems (password and biometric feature based) use one-time static authentication methods. Such systems are susceptible to masquerade attacks, where unauthorized users can take over a user's identity after the initial authorization of the user. A real time continuous authentication system provides better security control where the user is continuously authenticated based on the user's behavior while using the system. Monitoring more of user's features have shown to yield more accurate results. In this research, for continuous authentication of smartphone users, we use micro movements, orientation, and the grasp of the user's hand as a set of behavioral features. These user attributes can be easily collected from smartphone sensors like the accelerometer and gyroscope. We demonstrate the use and effectiveness of wrapper based feature selection over the filter based feature selection method in hand motion based continuous authentication systems. We also show how the use of fuzzy one class SVM can improve the accuracy of such hand motion based continuous authentication systems.