DGM : A Data Generative Model to Improve Minority Classes Presence in Anomaly Detection Domain

This repository provides a Keras-Tensorflow implementation of an approach of generating artificial data to balance network Intrusion Benchmark datasets using Generative Adversarial Networks. Benchmarking datasets for Network Intrusion Detection : NLS-KDD and UNSW-NB15

Prerequisites

• Keras >= 2.0.8
• TensorFlow
• Numpy >= 1.13.3
• Matplotlib >= 2.0.2
• Seaborn >= 0.7.1
• tabulate 0.8.6
• imbalanced-learn
• category_encoders

All the libraries can be pip installed

Getting Started

1. Clone this repo (for help see this tutorial).
2. Navigate to repository folder
3. Install dependencies which are specified in requirements.txt. use pip install -r requirements.txt or pip3 install -r requirements.txt
4. Raw Data is being kept here within this repo.
5. Navigate to desired Data Generative Model cd NSL-KDD or cd UNSW-NB15, then train and test the model by running train and test scripts : python train.py first, then python test.py.

Repository directory layout

.
├── Data                    # Benchmark datasets folder
│   ├── NSL-KDD             # NLS-KDD Dataset folder
│   ├── UNSW-NB15           # UNSW-NB15 Dataset folder
│   └── README.md           # Dataset info
├── NSL-KDD                 # Implementation for NSL-KDD dataset
│   ├── models              # Directory with implementation of the Generative Adversarial Networks and ML Classifiers
│   ├── train.py            # Main file for testing model
│   ├── test.py             # Main file for testing model
│   ├── README.md           
│   └── ...
├── UNSW-NB15               # Implementation for UNSW-NB15 dataset
│   ├── models              # Directory with implementation of the Generative Adversarial Networks and ML Classifiers
│   ├── train.py            # Main file for training and testing model
│   ├── test.py             # Main file for testing model
│   ├── README.md
│   └── ...
└── README.md

Contributions

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change or improve.

Contact

If you would like to get in touch, please contact:
Gcinizwe Dlamini - [email protected]

Citations and Contact.

You find a PDF format here : DGM: a data generative model to improve minority class presence in anomaly detection domain

If you use our work, please also cite the paper:

@article{dlamini2021dgm,
  title={DGM: a data generative model to improve minority class presence in anomaly detection domain},
  author={Dlamini, Gcinizwe and Fahim, Muhammad},
  journal={Neural Computing and Applications},
  pages={1--12},
  year={2021},
  publisher={Springer}
}