DRL-GAN: A Hybrid Approach for Binary and Multiclass Network Intrusion Detection

Area of Science:

• Cybersecurity
• Machine Learning
• Network Security

Background:

• Network-based attacks are increasing globally.
• Intrusion Detection Systems (IDS) are crucial for network security.
• Existing Machine Learning-based IDSs struggle with detecting uncommon attack types due to imbalanced datasets.

Purpose of the Study:

• To develop a novel hybrid technique for enhancing Intrusion Detection Systems (IDS).
• To improve the classification of minority attack classes in network traffic data.
• To address the limitations of current Machine Learning-based IDSs in handling imbalanced datasets.

Main Methods:

• Implemented a hybrid approach combining Generative Adversarial Networks (GAN) and Deep Reinforcement Learning (DRL).
• Trained a GAN model on the NSL-KDD dataset to generate synthetic network traffic data.
• Utilized the synthetic data to train a DRL model for intrusion detection.

Main Results:

• The hybrid GAN-DRL model demonstrated improved performance in classifying minority classes.
• Training the DRL model on GAN-generated synthetic data outperformed training on the original imbalanced dataset.
• Enhanced detection and classification accuracy for less frequent network attack types were observed.

Conclusions:

• Synthetic data generation using GANs can significantly improve the effectiveness of DRL-based IDSs.
• This hybrid approach offers a promising solution for detecting rare and challenging network attacks.
• The findings highlight the potential of GANs in addressing data imbalance issues in cybersecurity.