With large-scale Internet-connected systems, cyber-attacks often occur among the devices. The complexity and dynamic nature of cyber-attacks require protecting mechanisms to be responsive, adaptive, and scalable. The existing intrusion detection system detects several attacks based on the inspection of network traffics. It also protects and reduces the network from malicious activities. ConventionalMachine learning deep learning methods have introduced different techniques to detect malicious activities from intruders. However, it does not effectively consider the resource allocation constraints in a cloud environment and fail to learn and detect the unknown or new anomalies due to the behavior variations among the attacks. Moreover, detecting several attacks has become a crucial task in the cloud environment. Thus, deep reinforcement learning (DRL) methods enable highly capable of solving complex, dynamic, and high-dimensional cyber defense problems and intrusion detection. It effectively detects the new anomalies and solves the scalability problems caused by existing deep reinforcement learning.