Adaptive, ℇ-greedy reinforcement learning, IoT-Edge computing, Network security, Malware attacks, Task scheduling, Computer engineering. Computer hardware, TK7885-7895, Computer software, QA76.75-76.765

Abstract Attacks on IoT devices are increasing day by day. Since IoT devices nowadays have become an integral part of our daily lives, the data gathered from IoT devices benefits intruders in many ways. Financial and Healthcare institutions also allow their customers to use their services by using handheld IoT devices. Smart homes and autonomous vehicles use many IoT devices to gather data through the built-in sensors and send it to the Edge server for further processing. The computation result on the Edge server determines the decision to fulfill the user-specific needs. As these data are vital in the future cycle of execution of an intelligent algorithm of IoT device software program, hence the data are not just of temporary use, but it is transferred to a Cloud server for permanent storage. The data flows from IoT sensors to the Edge server, then from the Edge server to the Cloud server. Here the riskiest part for data to stay is on the Edge server. To counter such a security risk, we proposed and implemented the Adaptive Epsilon Greedy Reinforcement Learning (AEGRL) method which is the extension of the traditional Epsilon (ℇ) greedy reinforcement learning method. The proposed method works efficiently for both static and dynamic environments. Experimental results show that our proposed security method outperforms the recent similar security approaches in terms of scalability, robustness, and accuracy.