A Deep Reinforcement Learning-Based Attack Detection Model: G-RXAD

Cyber attacks are a growing concern for network services, especially with the rapid advancement of online platforms. Innovative technologies and methods are continuously being developed to counteract these evolving cyber threats and improve intrusion detection systems (IDS). Reinforcement learning techniques, including Markov algorithms and other systematic reinforcement learning algorithms, have emerged as promising tools for enhancing network security. However, the current method still has some shortcomings. This study proposes the G-RXAD model, which combines recursive feature elimination (RFE), adaptive synthetic sampling (ADASYN), and deep reinforcement learning (DRL). The proposed model can effectively identify a wide range of network attacks, significantly reduce irrelevant data, and address data imbalance. The effectiveness of this model was validated using the CSE-CIC-IDS2018 dataset, a comprehensive collection of genuine network traffic scenarios. The results indicate that the G-RXAD model can detect network threats more efficiently than conventional machine learning approaches. The proposed model excels in detecting network threats with fewer features, showcasing its potential for application in unmanned network environments and its superior performance in unmanned IDS systems across complex network environments.