https://github.com/aryanxxvii/lstm-ddos-prediction

LSTM neural network in PyTorch to predict DDoS attacks from network packet data, with 95% accuracy
https://github.com/aryanxxvii/lstm-ddos-prediction

ddos kafka network-analysis pyspark pytorch

Last synced: 7 days ago
JSON representation

LSTM neural network in PyTorch to predict DDoS attacks from network packet data, with 95% accuracy

• Host: GitHub
• URL: https://github.com/aryanxxvii/lstm-ddos-prediction
• Owner: aryanxxvii
• Created: 2024-08-18T17:21:51.000Z (6 months ago)
• Default Branch: main
• Last Pushed: 2024-11-11T10:40:12.000Z (3 months ago)
• Last Synced: 2024-11-11T11:36:02.298Z (3 months ago)
• Topics: ddos, kafka, network-analysis, pyspark, pytorch
• Language: Python
• Homepage:
• Size: 2.18 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
## DDoS Attack Detection with Stacked-LSTM and Real-Time Inference

Developed a 4-layer stacked-LSTM model to detect DDoS attacks, achieving **95.94% accuracy** on the CIC-DDoS2019 dataset using PyTorch and Numpy. The project involved:

![Connections per Second](_resources/connections_per_time.png)  

- **Data Processing:** 

  - Utilized **PySpark** to process and combine data from 11 different network packet sources into a single NumPy object for training and testing.

  

- **Model Development:**

  - Implemented a 4-layer stacked-LSTM model in **PyTorch** for time-series classification to differentiate between benign and DDoS attack traffic.

- **Real-Time Inference with Apache Kafka:**

  - Created a **Kafka Producer** to simulate real-time network packet flow using test data.

  - Developed a **Kafka Consumer** that takes in packets in real-time, performs inference with the trained LSTM model, and classifies them as DDoS or benign traffic.

- **Evaluation**:

    The performance of the model was evaluated using various metrics and visualizations to assess its ability to detect DDoS attacks accurately:

    - **Accuracy:** 95.94%

    - **Precision:** 98.21% 

    - **Recall:** 97.24%  

    - **F1 Score:** 97.73%

  

### Technologies Used:

- **PyTorch**: For building and training the stacked-LSTM model.

- **Kafka**: For simulating real-time packet flow (Producer and Consumer).

- **PySpark**: For data preprocessing and combining multiple data sources.

- **Seaborn/Matplotlib**: For data visualization.

- **NumPy**: For handling and manipulating data.