Gan Balancing: Towson University COSC 490 Research Data

Developed by Towson-Research for COSC 490 at Towson University, this repository focuses on Generative Adversarial Networks (GANs) for balancing the KDD99 intrusion detection dataset. Last updated in July 2021, the project utilizes TensorFlow and Keras to address class imbalance through synthetic data generation.

Use Cases

• Training GAN models using TensorFlow and Keras to generate synthetic network traffic features
• Evaluating class balancing techniques on KDD99 attack labels to improve classifier performance
• Testing MySQL-based data ingestion pipelines for machine learning workflows

Strengths

• Utilizes the KDD99 benchmark, a standard reference for network intrusion research
• Integrates GAN-based synthetic data generation to address minority class representation
• Developed within an academic research framework at Towson University

Limitations

• KDD99 is known to contain significant numbers of redundant and duplicate records
• Lack of explicit documentation regarding the specific GAN architecture used for generation
• Temporal staleness as the underlying network traffic patterns reflect 1999-era environments

Provenance

Source

Towson University Research (COSC 490)

Collection Method

Synthetic data generation via GANs applied to the KDD99 dataset

Freshness

The dataset and associated code have not been updated since July 15, 2021.