Deep Reinforcement Learning-Enhanced Intrusion Detection System for Cyber Threat Mitigation
Judy Simon, Nellore Kapileswar, S Diyananthan, A. M., A Hariprasath · 2025
Today cybersecurity remains a complex issue since new, more advanced attacks on networks occur regularly. In this paper, we design and present a new concept of Deep Reinforcement Learning-Enhanced Intrusion Detection System known as DRL-IDS, which effectively predicts and prevents cyber threats with satisfactory performance. The DRL-IDS uses deep learning integration of CNN for feature extraction and reinforcement learning for adaptive improvement of decision making in intrusion detection. Here, it is proposed that multi-layered network traffic data preprocessing and advance dynamic feature selection will enhance the ability to identify anomalies. prominent contributions are the adaptive policy for the accuracy of detection and the computational overhead and the novel reward that addresses the trade-off between false positive rate and the time taken to detect. Algorithms on benchmark cyber security datasets reveal enhanced detection efficiency and accuracy (approx 97.8 %), increased efficacy in terms of low false positives and high load handling capacity. These conclusions confirm DRL-IDS as a promising prospect for being an effective solution for today’s threats.