An Efficient Framework for Continuous Non Intrusive User Authentication Using Deep Learning

Abstract

Authentication is the first step in keeping personal information confidential in the newlinedigital world. The user authentication processes verify the legitimacy of users by newlinemethods like login credential verification, biometrics, voice recognition, digital newlinesignature and so on. Traditionally, one-time login based credential verification has newlinemany limitations and is prone to cyber attacks. Furthermore, single authentication at newlinethe start of a session leaves a user open to intrusion if the user leaves the station newlineunattended. Several new approaches are proposed to enhance the user authentication newlineframework. They are found to be ineffective and inconsistent. In recent years, newlinecontinuous user authentication (CUA) based on mouse and keystroke dynamics have newlinebeen studied extensively. They have inherent advantages like non intrusiveness, non newlinerequirement of additional hardware and improved security. A commercially deployable newlineCUA system for multiple application areas is yet to be seen because fast authentication newlineand high performance are yet to be attained. In addition, security issues have only been newlinecompounded by the advent of distributed networks and global internet availability. Combating these issues require an understanding of the mouse dynamics and newlinekeystroke features that discriminate users effectively. Input optimization for improved newlineperformance is an essential precondition to build an efficient machine learning model. newlineSince performance and speed depends on many factors like an optimal input search newlinespace, discriminative feature selection and effectiveness of the learning algorithm for newlinethe given environment, makes the problem challenging. newline