Improving Intrusion Detection Systems with Multi-Agent Deep Reinforcement Learning: Enhanced Centralized and Decentralized Approaches

Intrusion detection is a crucial task in the field of computer security as it helps protect these systems against malicious attacks. New techniques have been developed to cope with the increasing complexity of computer systems and the constantly evolving threats. Multi-agent reinforcement learning (MARL), is an extension of Reinforcement Learning (RL) in which agents can learn to detect and respond to intrusions while considering the actions and decisions of the other agents. In this study, we evaluate MARL's performance in detecting network intrusions using the NSL-KDD dataset. We propose two approaches, centralized and decentralized, namely COCA-MADQN and MADQN-GTN. Our approaches show good results in terms of Accuracy, Precision, Recall, and F1-score.