onuion

<br>

Onuion

Real-time Security Risk Analysis Model

Open-source hybrid (rule-based + ML) security risk analysis system for session data analysis with sub-millisecond inference time.

🎯 Purpose

onuion is a production-ready risk analysis system that detects security risks by analyzing session data. Using a hybrid approach (rule-based + ML), it provides both fast detection with deterministic rules and sophisticated pattern recognition with machine learning.

⚡ Features

• Real-time Inference: < 1ms inference time target
• Hybrid System: Rule-based + TensorFlow ML model
• Small Model: ~2,000 parameters, optimized for tabular data
• Production-Ready: Modular, tested, documented code
• Open Source: Licensed under MIT

🏗️ Architecture

Session Data
    ↓
[Feature Extractor] → 25 numeric/categorical features
    ↓
[Rule Engine] → Deterministic risk detection (< 0.1ms)
    ↓
[TensorFlow Model] → ML-based risk probability (< 0.5ms)
    ↓
[Risk Aggregator] → Final riskScore + risk array
    ↓
Risk Analysis Result

Components

1. Feature Extractor: Converts session data into 25 features

   • IP change metrics
   • Geo-location differences
   • Device fingerprint changes
   • Request rates and patterns
   • Session timing information

2. Rule Engine: Fast risk detection with deterministic rules

   • ip_mismatch: IP change
   • session_hijacking: Session hijacking indicators
   • bot_behavior: Bot behavior patterns
   • geo_anomaly: Geo-location anomaly
   • device_fingerprint_mismatch: Device fingerprint mismatch
   • rapid_ip_change: Rapid IP change
   • suspicious_request_pattern: Suspicious request patterns

3. TensorFlow Model: Feed Forward Neural Network

   • Input: 25 features
   • 3 Dense layers (64 → 32 → 16 units)
   • Output: Risk probability (0.0 - 1.0)
   • Total parameters: ~2,000

4. Risk Aggregator: Combines rule and ML scores

   • Weighted combination (Rule 40%, ML 60%)
   • Final riskScore (0-100)
   • Risk array generation

📦 Installation

Requirements

• Python 3.8+
• TensorFlow 2.10+
• NumPy 1.21+

Installation Steps

# Clone repository
git clone https://github.com/onuion/onuion.git
cd onuion

# Create virtual environment (recommended)
python -m venv venv
source venv/bin/activate  # Linux/Mac
# or
venv\Scripts\activate  # Windows

# Install dependencies
pip install -r requirements.txt

# Install in development mode (optional)
pip install -e .

🚀 Usage

Basic Usage

from onuion import analyze_risk

# Session data
session_data = {
    "current_ip": "192.168.1.100",
    "initial_ip": "192.168.1.50",
    "ip_history": ["192.168.1.50", "192.168.1.100"],
    "current_geo": {"country": "TR", "city": "Istanbul"},
    "initial_geo": {"country": "TR", "city": "Ankara"},
    "current_device": {"fingerprint": "fp123"},
    "initial_device": {"fingerprint": "fp123"},
    "current_browser": {},
    "initial_browser": {},
    "requests": [
        {"timestamp": 1706000000, "method": "GET", "endpoint": "/api/users"}
    ],
    "session_duration_seconds": 10.0,
    "current_session_id": "sess_123",
    "initial_session_id": "sess_123",
    "current_cookies": {},
    "initial_cookies": {},
    "current_referrer": "",
    "initial_referrer": ""
}

# Risk analysis
result = analyze_risk(session_data)

# Results
print(f"Risk Score: {result.riskScore}/100")
print(f"Risks: {result.risk}")
print(f"Inference Time: {result.inference_time_ms:.3f} ms")

Output Format

{
    "riskScore": 45.2,  # Risk score (0-100)
    "risk": ["ip_mismatch", "geo_anomaly"],  # Detected risk types
    "rule_score": 40.0,  # Rule engine score
    "ml_score": 48.0,  # ML model score
    "confidence": 92.0,  # Confidence score
    "inference_time_ms": 0.856  # Inference time (ms)
}

Examples

See the examples/ directory for more examples:

• examples/basic_usage.py: Basic usage example
• examples/load_from_json.py: Load from JSON file
• examples/sample_session.json: Example session data

🧪 Model Training

Training with Synthetic Data (for testing)

python -m onuion.train --synthetic --epochs 50 --output-dir models/onuion_model

Training with Real Data

# Data format: .npz file
# Contents: X_train, y_train, X_val, y_val (numpy arrays)

python -m onuion.train --data-path data/training_data.npz --epochs 100

Hugging Face Integration

onuion supports Hugging Face Hub for model sharing and deployment.

Model Hub: https://huggingface.co/onuion/onuion

Upload a model to Hugging Face Hub:

from onuion.huggingface import upload_to_hub

# Upload a saved model
upload_to_hub(
    model_path="models/onuion_model",
    repo_id="onuion/onuion",
    token="your_hf_token",  # Optional, can use huggingface-cli login
    private=False
)

Download a model from Hugging Face Hub:

from onuion.huggingface import download_from_hub
from onuion import analyze_risk

# Download model
model = download_from_hub(
    repo_id="onuion/onuion",
    local_dir="models/downloaded_model"
)

# Use with inference
model.save("models/downloaded_model")
result = analyze_risk(session_data, model_path="models/downloaded_model")

Install Hugging Face dependencies:

pip install huggingface_hub

See examples/huggingface_upload.py and examples/huggingface_download.py for complete examples.

📊 Performance

Inference Time

• Target: < 1ms
• Average: ~0.5-0.8ms (CPU)
• P95: < 1.5ms
• Throughput: ~1,000-2,000 requests/second

Benchmark

python benchmark/benchmark.py

🧪 Testing

# Run all tests
pytest tests/

# With coverage
pytest tests/ --cov=onuion --cov-report=html

📊 Diagrams

📚 Documentation

See the docs/ directory for detailed documentation:

• docs/model.md: Model architecture details
• docs/risk_types.md: Risk type descriptions
• docs/api.md: API reference

🤝 Contributing

We welcome your contributions! Please:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

📝 License

This project is licensed under the MIT License. See the LICENSE file for details.

🙏 Acknowledgments

• TensorFlow team
• Open source community
• All contributors

📧 Contact

You can open an issue or send a pull request for questions.

---

Note: This project is prepared for production use, but it is recommended to use a model trained with real data. The default model is only for testing inference purposes.