In the digital age, frightening patterns in digital threats are emerging. It is impossible to ignore threats to IoT networks. Threats can take on any of the typical forms, including Denial-of-Service (DoS), Distributed Denial-of-Service (DDoS), Virus assault, Man-in-the-middle attack (Mitm), Advanced Persistent Threats (APT), Password Assault, and more. It is crucial to eliminate all threats from IoT networks and devices. Reinforcement learning to detect anomalies in an IoT network is seen to be the greatest option for correcting risks in a network, hence fixing the afflicted nodes, according to this thesis, "Self-Adaptation of SDN-based IoT Networks." (Markov) MDP policies and MAPE-K loop properties in Self-aware systems are the bases of the design in this thesis. The network system exhibited self-adaptability features, which makes it self-correcting and self-healing. The objective of this research is to propose a means to secure the devices in an IoT network by protecting them from any form of threats and ensuring that the devices function normally. Even at the advent of abnormal functioning of any node in the network, the system should be able to correct itself. A Software Defined Network (SDN) architecture is proposed for the design in a later section, which explains the kind of SDN that should be in place for the intrusion detection system. Further into the thesis, we dived deep into the general overview of deep reinforcement learning. Then comes the implementation, which talks about the kind of reinforcement learning policy used in the work and how the result was derived. The other section discusses the result and discussion, where the result in this work was compared with the result of the traditional machine learning algorithm.