Analysis of Anomalous Behavior in Network Systems Using Deep Reinforcement Learning with CNN Architecture

—In order to gain access to networks, differ- ent types of intrusion attacks have been designed and improved. Computer networks have become increasingly important in daily life due to the increasing reliance on them. In light of this, it is quite evident that algorithms with high detection accuracy and reliability are needed for various attack types. The purpose of this paper is to develop an intrusion detection system that is based on deep reinforcement learning. Based on the Markov decision process, the proposed system can generate informative representations suitable for classification tasks based on vast data. This paper inspects reinforcement learning from two perspectives: deep Q learning and double deep Q learn- ing. Different experiments have demonstrated that the proposed systems have an accuracy of 99 . 17% over the UNSW-NB15 dataset in both approaches, an improvement over previous methods based on contrastive learning and LSTM-Autoencoders. The performance of the model trained on UNSW-NB15 has also been evaluated on BoT-IoT dataset, resulting in competitive performance.