Credit Risk Prediction System (MLOps)

A production-grade end-to-end Machine Learning project for predicting credit default risk, built with modern MLOps practices.
The system covers the full ML lifecycle: data processing, model training, experiment tracking, model registry, and API-based deployment.

---

📌 Project Overview

This project aims to predict whether a loan applicant is likely to default based on demographic, financial, and credit history features.
Multiple models were trained and evaluated, with the best-performing model deployed as a REST API.

---

🧠 Key Features

• End-to-end ML pipeline (data → model → API)
• Feature engineering and preprocessing
• Model comparison and evaluation
• Experiment tracking and model versioning
• Production-ready inference API

---

🛠️ Tech Stack & Skills

• Programming: Python
• Data Processing: Pandas, NumPy
• Machine Learning: Scikit-learn, XGBoost
• MLOps: MLflow (tracking, registry)
• API: FastAPI, Uvicorn
• Evaluation: ROC-AUC, Precision, Recall

---

📊 Models Used

• Logistic Regression (baseline)
• XGBoost Classifier (final model)

XGBoost achieved higher recall (~71%), making it more effective at identifying high-risk borrowers.

---

🧪 Experiment Tracking

MLflow is used to:

• Track experiments and metrics
• Compare multiple models
• Register the best-performing model

Note: MLflow artifacts (mlruns/) are generated locally at runtime and are excluded from version control.

---

🚀 Project Structure

Credit_risk_2.0/ │ ├── src/ │ ├── data/ # Data loading & preprocessing │ ├── features/ # Feature engineering │ ├── models/ # Training & evaluation scripts │ └── api/ # FastAPI inference service │ ├── notebooks/ # Exploration & experiments ├── requirements.txt ├── README.md └── .gitignore

---

🔮 Running the Project

1. Install dependencies

pip install -r requirements.txt
2. Train models & log experiments
python src/models/train.py
3. Start the API server
uvicorn src.api.app:app --reload
4. Test the API
Open your browser at:

http://127.0.0.1:8000/docs
🔑 Key Insight
Gradient-boosted models captured non-linear patterns in credit data better than linear models, significantly improving recall for default prediction — a critical metric in real-world credit risk systems.

📈 Future Improvements
Data drift and model monitoring

Automated retraining pipelines

CI/CD for ML workflows

Dockerized deployment

👤 Author
Fardeen
Aspiring AI/ML Engineer | Interested in applied ML, MLOps, and data-driven systems