Trustworthy Communication Channel for the IoT Sensor Nodes Using Reinforcement Learning

Abstract

IoT has been deployed in different fields to enhance the quality of human life. However, the IoT has become an appealing source for intruders to penetrate the smart premises of users. As security technology grows, cybercriminals also enable themselves to launch the most sophisticated attacks. Therefore, to maintain the protection of IoT devices, there is need for a responsive security system that can efficiently encounter novel attacks. This paper proposes a security mechanism to tackle cyberattacks by employing Reinforcement Learning (RL). Through RL, we can efficiently detect any ordinary or novel attacks as the RL agent learns by its own without human instructions. So, it educates the algorithm against any sophisticated attack. Dataset UNSW-NB is incorporated to evaluate the performance of the proposed study. The performance and detection rate of the model was enhanced selecting optimal features of the dataset. The proposed RL approach achieves an average accuracy of 97.29%. Results reveal that the proposed study has the potential to be deployed as a security mechanism against cyberattacks.