A novel multi-module integrated intrusion detection system for high-dimensional imbalanced data.

Area of Science:

• Cybersecurity
• Network Intrusion Detection
• Machine Learning

Background:

• Network data presents challenges due to high dimensionality, complexity, and imbalance.
• Existing intrusion detection systems struggle with minority class detection accuracy, identifying novel threats, and minimizing false alarms.

Purpose of the Study:

• To propose a novel multi-module integrated intrusion detection system (GMM-WGAN-IDS).
• To enhance the detection of minority classes, unknown attacks, and reduce false alarm rates in network intrusion detection.

Main Methods:

• Feature extraction using a stacked autoencoder (SAE) module for deeper data representation.
• Imbalance processing via a Gaussian Mixture Model (GMM) clustering and Wasserstein Generative Adversarial Network (WGAN) hybrid (GMM-WGAN).
• Classification using a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) hybrid (CNN-LSTM).

Main Results:

• The GMM-WGAN-IDS demonstrated superior performance on the NSL-KDD and UNSW-NB15 datasets.
• The proposed system outperformed state-of-the-art intrusion detection methods in experimental evaluations.

Conclusions:

• The GMM-WGAN-IDS effectively addresses key challenges in network intrusion detection.
• The integrated multi-module approach offers significant improvements in accuracy and detection capabilities for network security.