A Hybrid Reinforcement Model Using Deep Q-Learning for Attack Detection

Computer network security can be ensured by timely detection of unauthorized access and impending attacks. In order to identify computer network attacks, this paper introduces a hybrid model for creating an intrusion detection system. The proposed model uses in-depth reinforcement learning to create an intelligent agent with a high understanding of the data being transmitted over the network to be able to detect network attacks well. The proposed model also uses PCA to represent new data because agent training is highly dependent on input data. The NSL-KDD dataset, besides the CTU-13 dataset, has been used as a standard dataset to train and test the proposed model, and in the training phase, an attempt has been made to overcome its challenges. The results show the appropriate accuracy and fall alarm rate for the proposed model in the study dataset.