Intrusion Detection Dashboard

Overview

An interactive Streamlit app for visualizing network session data and predicting cyberattacks using a LightGBM model. The model backend is served via a Flask API hosted on Hugging Face Spaces. Enables real-time intrusion prediction based on session characteristics, interactive filtering and visualization of network traffic, and an API-backed model inference using a threshold-optimized LightGBM classifier.

Model Details

• Model: LightGBM Classifier
• Threshold: 0.2 (favoring recall over precision)
• Recall: 87.1%
• Precision: 62.5%
• F1 Score: 73.0%

Features

• Interactive Prediction: Input session data and receive real-time attack predictions.
• Visual Insights: Explore traffic patterns by protocol, encryption type, and more.
• API Integration: Live inference powered by a deployed Flask API.
• Custom Threshold: Lowered to detect more attacks at the expense of precision.

Links

• Try the Dashboard
• Model Training Notebook
• Dataset on Kaggle

Example API Request

curl -X POST https://e-eeeema-intrusion-detection.hf.space/predict \
     -H "Content-Type: application/json" \
     -d '{"features": [500, 3, 500.0, 0.5, 1, 0, 1, 0, 0, 1]}'

License

MIT