An Imbalanced Generative Adversarial Network-Based Approach for Network Intrusion Detection in an Imbalanced Dataset

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Network Intrusion Detection Systems (NIDS) are crucial for network security.
• Class imbalance in datasets limits the effectiveness of NIDS, particularly for minority attack classes.
• Existing methods struggle to accurately detect low-frequency, high-impact cyber threats.

Purpose of the Study:

• To propose an Imbalanced Generative Adversarial Network (IGAN) to address class imbalance in NIDS datasets.
• To improve the detection rate of minority attack classes without compromising overall system efficiency.
• To enhance the accuracy and reliability of network security monitoring.

Main Methods:

• Data preprocessing including normalization and one-hot encoding.
• Utilizing IGAN for oversampling minority class samples to balance the dataset.
• Employing an ensemble model of LeNet 5 and Long Short-Term Memory (LSTM) for classification of network traffic.
• Evaluating performance using metrics like accuracy, precision, recall, TPR, FPR, and F1-score.

Main Results:

• The proposed IGAN approach significantly improved the detection rate of minority attack classes.
• The ensemble LeNet 5 and LSTM model achieved over 98% accuracy in classifying network attacks.
• IGAN oversampling enhanced overall detection accuracy and performance metrics compared to other deep learning methods.
• The method demonstrated superior precision, recall, TPR, FPR, and F1-score.

Conclusions:

• IGAN effectively mitigates class imbalance issues in NIDS datasets.
• The proposed method offers a robust solution for improving the detection of diverse cyber threats.
• IGAN combined with ensemble deep learning provides state-of-the-art performance in network intrusion detection.