AI Based Intrusion Detection System (IDS)

• Implementation of an NLP-based Intrusion Detection System (IDS) for binary classification of detected attack packets.
• This task won 1st place (과기정통부 장관상) in the Cybersecurity AI Big Data Challenge (Nov 2022).

📋 Task

The primary task is to classify intrusion detection system (IDS) results into attack packet or non-attack packet, using a binary classification approach.

🤖 Model

• Base Model: RoBERTa (SecureBERT)
  • Fine-tuned on IDS-related binary classification data.
  • Leverages pre-trained language model capabilities for analyzing attack packet data.

📊 Dataset

• Intrusion Detection System Dataset
  • Contains labeled samples for binary classification.
  • Size: N million samples.
  • Includes features extracted from network traffic packets:

  'PAYLOAD', 'APP_PROTO', 'SRC_PORT', 'DST_PORT', 'IMPACT', 'RISK', 'JUDGEMENT', 'Method', 'Method-URL', 'HTTP', 'Host', 'User-Agent', 'Accept', 'Accept-Encoding', 'Accept-Language', 'Accept-Charset', 'Content-Type', 'Content-Length', 'Connection', 'Cookie', 'Upgrade-Insecure-Requests', 'Pragma', 'Cache-Control', 'Body'

📂 Repository Structure

IDS-BERT/
├── ckpt/                 
│   ├── pretrained/              
│   └── trained/                 
├── dataset/                    
├── data_preprocess.py        
├── train.py    
├── inference.py          
├── utils.py                 
├── main.ipynb                       
└── README.md                   

📚 Requirements

• Python: 3.9+
• CUDA: 11.7+ (for GPU-based training and inference)
• For a complete list of dependencies, see requirements.txt.

🚀 Getting Started

1. Clone the Repository

git clone https://github.com/cv-lee/IDS-BERT.git

2. Prepare the Dataset & Pretrained Model

• Place your dataset files in the dataset/ folder.
• Place your pretrained files in the ckpt/pretrained folder.
• Open the main.ipynb file.
• Execute the data preprocessing step:

  python3 data_preprocess.py

3. Train the Model

• Train the RoBERTa model using the preprocessed dataset and pretrained model:

  python3 train.py

4. Run Inference

• Use the trained model for binary classification:

  python3 inference.py

📄 Configuration

{
  "max_seq_length": 512,
  "batch_size": 16,
  "learning_rate": 1e-5,
  "weight_decay": 0.01,
  "num_epochs": 10,
  "device": "cuda"
}

📬 Contact

For any questions or issues, please contact:

• Joohyun Lee