CTTGAN: Traffic Data Synthesizing Scheme Based on Conditional GAN.

Area of Science:

• Cybersecurity
• Machine Learning
• Data Science

Background:

• Machine learning algorithms struggle with imbalanced datasets, common in malicious traffic identification.
• Benign network traffic vastly outweighs malicious traffic, leading to poor detection rates for rare malicious samples.

Purpose of the Study:

• To present a novel model, Conditional Tabular Traffic Generative Adversarial Network (CTTGAN), for synthesizing traffic samples.
• To balance imbalanced network traffic datasets and enhance the identification rate of malicious traffic.

Main Methods:

• Utilized the Conditional Tabular Generative Adversarial Network (CTGAN) algorithm for data synthesis.
• Expanded small-category traffic samples to create a balanced dataset for machine learning.
• Processed both discrete and continuous variables in traffic feature data simultaneously.

Main Results:

• Achieved a recognition rate exceeding 0.99 for expanded samples across MLP, KNN, and SVM algorithms.
• Demonstrated superior performance of the CTTGAN model compared to traditional oversampling and undersampling techniques.
• Reduced storage costs and computational complexity relative to image-based data models.

Conclusions:

• CTTGAN effectively balances imbalanced network traffic datasets by synthesizing minority class samples.
• The proposed model improves the accuracy of malicious traffic identification, offering a practical solution for cybersecurity.
• CTTGAN efficiently handles mixed data types and maintains data distribution integrity.