Credit Score API

FastAPI REST API for credit scoring predictions deployed on Google Cloud Run.

Overview

Production-ready API that provides credit risk predictions using a LightGBM model with preprocessing pipeline. Returns probability of payment difficulties for loan applicants.

Live API: https://credit-score-api-572900860091.europe-west1.run.app

Features

• RESTful API with FastAPI
• Unified sklearn pipeline (preprocessing + model)
• Automatic input validation
• CORS enabled for dashboard integration
• Deployed on Google Cloud Run (serverless)

API Endpoint

POST /predict

Request:

{
  "id": 162473
}

Response:

{
  "credit_score": 0.215,
  "advice": "No payment difficulties"
}

Status Codes:

• 200: Success
• 404: Client ID not found
• 500: Internal server error

Project Structure

P7_api_deploy/
├── app/
│   ├── main.py              # FastAPI application
│   ├── models/              # Pydantic schemas
│   ├── services/            # Business logic (CreditScorer)
│   └── api/                 # API routes
├── ressource/
│   └── pipeline.joblib      # Unified sklearn pipeline (5.2 MB)
├── data/
│   └── dataset_sample.csv   # Sample client data
├── Dockerfile               # Container configuration
└── requirements.txt         # Python dependencies

Local Development

Installation

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # or `.venv\Scripts\activate` on Windows

# Install dependencies
pip install -r requirements.txt

Run Locally

# Start development server
uvicorn app.main:app --reload --port 8001

# Test the API
curl -X POST "http://127.0.0.1:8001/predict" \
  -H "Content-Type: application/json" \
  -d '{"id": 162473}'

Deployment

Deploy to Cloud Run

# Deploy from source
gcloud run deploy credit-score-api \
  --source . \
  --region europe-west1 \
  --allow-unauthenticated

# Test deployed API
curl -X POST "https://credit-score-api-572900860091.europe-west1.run.app/predict" \
  -H "Content-Type: application/json" \
  -d '{"id": 162473}'

Model Pipeline

The API uses a unified sklearn pipeline that includes:

1. Preprocessing: Imputation, scaling, encoding
2. Feature Selection: SelectFromModel with LightGBM
3. Classification: LightGBM with optimized hyperparameters

Decision Threshold: 0.47 (optimized for G-mean)

Tech Stack

• Framework: FastAPI 0.104.1
• ML: scikit-learn 1.6.1, LightGBM 4.6.0
• Server: Uvicorn
• Deployment: Google Cloud Run
• Container: Docker

Related Projects

• ML Training: P7_Implementez_modele_scoring - Model training and pipeline creation
• Dashboard: P7_dashboard_deploy - Streamlit dashboard consuming this API

API Documentation

Interactive documentation available at:

• Swagger UI: https://credit-score-api-572900860091.europe-west1.run.app/docs
• ReDoc: https://credit-score-api-572900860091.europe-west1.run.app/redoc

License

OpenClassrooms project - Educational purposes

---

Last update: November 2025