Enhancing Network Intrusion Detection Using Deep Reinforcement Learning: An Adaptive Learning Approach

In recent years due to the emergence of a wide variety of technologies like the Internet of Things and Cloud-based platforms the network become more vulnerable to security attacks. Several techniques and solutions have been implemented to resolve this issue, but the major problem in all of those methods is their inability to correctly detect the attack traffic since the dataset used has more data about normal traffic than the attack data. We propose a novel approach for network intrusion detection using deep reinforcement learning techniques on the NSL-KDD dataset. Traditional intrusion detection systems rely on predefined rules and signatures to identify malicious traffic, which may not be effective against new and sophisticated attacks. Our proposed method utilizes deep reinforcement learning algorithms to learn and adapt to new types of attacks in real time. We evaluate the performance of our model on the NSL-KDD dataset and compare it with other traditional machine learning-based intrusion detection methods. The results demonstrate that our proposed approach outperforms existing methods and achieves high accuracy and low false-positive rates, making it a promising solution for detecting network intrusions in real-world scenarios. Our work contributes to the advancement of the field of intrusion detection and provides valuable insight into the potential of deep reinforcement learning for enhancing network security. We were able to obtain a test accuracy of 81%, whereas the accuracy of the existing work which uses deep Q learning is about 78%. Our proposed work outperforms the existing work not only in terms of accuracy but also based on the capability to correctly classify the minority class.