EAARL: Enhanced Adaptive Authentication based on Reinforcement Learning

Secure authentication is a fundamental problem in any network. The size of networks within homes is rapidly increasing with the number of devices each person is now using. There is a need to develop a secure authentication technique that is versatile to different types of devices. In this paper, we review and analyze authentication techniques using reinforcement learning. Three techniques are reviewed in detail. The application area of these techniques is online money transactions. There are some issues with the existing methods, such as a lack of data confidentiality and vulnerability to insider attacks. To tackle these problems, we propose EAARL, Enhanced Adaptive Authentication based on Reinforcement Learning, implement it for an IoT application and provides a novel solution to the abovementioned problems. We use elliptic curve cryptography and reinforcement learning, which makes this solution unique. The implementation of this scheme is performed using Jupyter notebooks and Python 3.7.