Abstract
Most smartphone users prefer easy and convenient authentication without remembering complicated passwords or drawing intricate patterns. Preferably, after one-time authentication, there is no verification of the user's authenticity. Therefore, security and privacy against unauthorized users is a crucial research area. Behavioral authentication is an emerging security technique that is gaining attention for its uniqueness and transparency. In this paper, a behavior-based authentication system is built using swipe movements to continuously authenticate the user after one-time traditional authentication. The key feature is the selection of an optimal feature set for the swipe movement. Five machine learning classifiers are used, of which random forest is selected based on the best values of accuracy and F-measure. A real-time system is developed by shifting all of the computational power to a cloud server to overcome the smartphone's computational limitations. The system is tested on three smartphones, and it is found that a minimum of seven swipes is sufficient to check user authenticity. In our experiments, the proposed feature set performs better than a state-of-the-art feature set.