DefinePK

Abstract

Over the years, mobile device has become the most preferred way to communication, sharing information and even sensitive data. In fact, many security measurements such as 2-step verification, OTP based transaction are also dependent on the authentication of mobile user. A lot of research initiatives have tried to develop both explicit and implicit authentication mechanism for the devices. But implicit authentication measures have gained much attention over the year, as they are more secure continuous measure that can be made undetectable. Majority of the previous studies on implicit authentication of mobile device are mainly focused on analyzing different data attributes in retrieving behavioral traits and using linear techniques to describe an individuals behavioral pattern. On top of that, almost all the previous studies ignores the consideration of analyzing how a user interact with a mobile environment as they are more focused on finding pattern within the retrieved data features. To our knowledge for the first time, this study analyze the user interaction with the mobile phone by mapping them into the Markov Decision Process (MDP) environment and later proposes a RNN agent based technique to extract user’s unique policy that could be used as biometric traits. The proposed technique is also able to extract a continually evolving user’s policy by incorporating data aggregation while ensuring authenticity. The comparative analysis of the proposed technique clearly indicates the suitability of this approach, as the proposed technique is able to outperform some of the well-known implicit authentication techniques and attain a recognition rate of around 94%.