Network Intrusion Detection System using Reinforcement learning

Abstract

Our research on the efficacy of deep reinforcement learning helps us comprehend the challenges encountered by NIDS (DRL). To find network anomalies, we suggest integrating Adversarial/Multi Agent Reinforcement Learning with Deep QLearning (AE-DQN). We compare our suggestions on the NSL-KDD dataset with the KDDTest+ dataset. In this article, we take a look at the difficulty of reducing an infinite number of possible categories down to only five. Our strategy yielded an overall F1 score of 79% and an accuracy of 80% across the board. Furthermore, our proposed method outperforms the Recurrent Neural Network (RNN) IDS (2) and the Adversarial Reinforcement Learning with SMOTE (AESMOTE) IDS in terms of the variety of assaults it can identify, as shown by its performance on the NSL-KDD dataset (3). Our major aim going forward is to enhance detection efficiency against different kinds of threats.