End-to-end machine learning pipeline for fraud detection with real-time simulation, data drift monitoring, and interactive dashboards.

breeboost/
├── src/
│   ├── inference.py            # Model inference logic
│   ├── utils/
│   │   └── logger.py           # Logging utility
│   └── ...
├── monitoring/
│   ├── reference.csv           # Clean data from training
│   ├── production.csv          # New data + predictions
│   ├── report.py               # Drift detection script
│   └── reports/
│       └── report.html         # Generated drift report
├── data/
│   └── processed/
│       └── paysim_cleaned.csv  # Cleaned dataset
├── app.py                      # Streamlit fraud dashboard
├── requirements.txt
└── README.md

flowchart TD
    A[Cleaned Data] --> B[Model Inference]
    B --> C[Predictions + Probabilities]
    A --> D[Reference Data]
    C --> E[Drift Detection]
    D --> E
    E --> F[HTML Report]

    style A fill:#f9f,stroke:#333,stroke-width:1px
    style B fill:#bbf,stroke:#333,stroke-width:1px
    style C fill:#bfb,stroke:#333,stroke-width:1px
    style D fill:#bbf,stroke:#333,stroke-width:1px
    style E fill:#ffb,stroke:#333,stroke-width:1px
    style F fill:#fc9,stroke:#333,stroke-width:1px

• Trained XGBoost classifier
• Input validation and prediction interface
• Outputs label + fraud probability

• Simulate transactions with input form
• View prediction results and probabilities
• Visualize important features and correlations

• Compares training vs production data
• Visualizes drift for numerical features
• Outputs full HTML diagnostics

🔗 View Latest Drift Report

pip install -r requirements.txt

python src/inference.py

python src/utils/extract_ref.py

Use the dashboard or inference module to generate rows for production.csv.

python monitoring/report.py

streamlit run app.py

• amount, oldbalanceOrg, newbalanceOrig
• errorBalanceOrig, errorBalanceDest
• hour, day, is_large_transaction

• Python 3.8+
• pandas, xgboost, joblib
• evidently, seaborn, matplotlib
• streamlit

Install everything:

pip install -r requirements.txt

• Performance drift monitoring (F1, recall)
• Notification triggers (e.g. Slack alerts)
• CI/CD with GitHub Actions
• Dockerization + cloud deployment
• Add feature importance explanation (e.g. SHAP)

• Run scripts from the project root for relative paths to resolve.
• Check monitoring/reports/report.html regularly to evaluate input stability.
• You can customize the model threshold and feature set in inference.py.

This project is licensed under the MIT License.