🕵️ Fraud Detection Pipeline

This project implements a machine learning pipeline for detecting fraudulent transactions using ensemble and distance-based classifiers. The pipeline includes preprocessing, model training, hyperparameter tuning, evaluation, and experiment tracking via MLflow.

---

📊 Models Compared

Two models were trained and evaluated:

• ExtraTreesClassifier
• KNeighborsClassifier

✅ Performance Summary

After cross-validation and evaluation on the test set:

• ExtraTreesClassifier consistently outperformed KNN in terms of:
  • F1-score
  • Recall
  • ROC AUC
• It also demonstrated better generalization and robustness across imbalanced class distributions.

📌 Conclusion: ExtraTreesClassifier is the recommended model for this fraud detection task.

---

📁 Project Structure

Fraud_detection/
├── src/                 # Source code modules
├── notebooks/           # Jupyter notebooks for exploration and prototyping
├── configs/
│   ├── config.yaml      # Data and path configurations
│   └── params.yaml      # Model parameters
├── .env                 # Environment variables (see below)
├── main.py              # Entry point for running the pipeline
├── requirements.txt     # Python dependencies
├── README.md            # Project documentation

---

🚀 Getting Started

Follow these steps to set up the environment and run the pipeline.

1. Clone the Repository

git clone https://github.com/Shahriyar-1988/Fraud_detection.git
cd Fraud_detection

2. Create and Activate a Virtual Environment

python -m venv venv
source venv/bin/activate       # On Windows: venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Configure Environment Variables

Create a .env file in the project root with the following content:

MLFLOW_TRACKING_URI=http://localhost:5000

⚠️ This URI points to the MLflow server used for experiment tracking. Make sure the server is running before executing the pipeline.

You can start a local MLflow UI using:

mlflow ui

Then open http://localhost:5000 in your browser.

5. Run the Pipeline

python main.py

---

🧪 Experiment Tracking with MLflow

This project uses MLflow for:

• Logging models and metrics
• Comparing model performance
• Storing artifacts like confusion matrices and classification reports

---

🤝 Contributing

Feel free to fork this repo and submit pull requests. Contributions for improving model performance, optimizing preprocessing, or enhancing logging are welcome!

---

📄 License

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