Loan Default Predictor API

A production-ready machine learning inference API that predicts whether a loan applicant is likely to default, along with the model’s confidence score.

The project demonstrates an end-to-end ML system: data → model → FastAPI → Docker → documented API.

---

🚀 What Problem This Solves

Loan providers need a quick, consistent, and explainable way to assess credit risk.

This API:

• Takes structured borrower information

• Predicts whether the borrower is likely to default

• Returns both:

  • a binary prediction
  • a probability score (risk confidence)

The API is designed to be:

• reproducible
• deployable
• easy to integrate into backend systems

---

🧠 Model Output Explained

The response contains two key fields:

Default

• 1 → borrower is predicted to default
• 0 → borrower is predicted to not default

This classification uses the 0.45 probability threshold.

---

probability

• A float between 0 and 1
• Represents the model’s confidence that the borrower will default

Example:

"probability": 0.72

Means:

“The model estimates a 72% chance of default.”

Returning probability allows risk thresholds to be adjusted later without retraining the model.

---

🛠️ How to Run (Docker – 2 Minutes)

1️⃣ Clone the repository

git clone https://github.com/Satadru03/fastapi-borrowing.git
cd loan-default-api

---

2️⃣ Build the Docker image

docker build -t loan-default-api .

---

3️⃣ Run the container

docker run -p 8000:8000 loan-default-api

---

4️⃣ Open API docs

Open in browser:

http://localhost:8000/docs

Interactive Swagger UI is available.

---

📡 API Endpoint

POST /predict

Accepts borrower information and returns a default prediction.

---

✅ Example – Normal Input (Low Risk)

Request

{
  "Age": 35,
  "Income": 60000,
  "LoanAmount": 12000,
  "CreditScore": 720,
  "MonthsEmployed": 48,
  "NumCreditLines": 3,
  "InterestRate": 8,
  "LoanTerm": 36,
  "DTIRatio": 0.25,
  "Education": "Bachelor's",
  "EmploymentType": "Full-time",
  "MaritalStatus": "Married",
  "LoanPurpose": "Auto",
  "HasMortgage": "No",
  "HasDependents": "No",
  "HasCoSigner": "No"
}

Response

{
  "Default": 0,
  "decision": "will_not_default",
  "probability": 0.345
}

---

⚠️ Example – Edge Case (High Risk)

Young borrower with minimal income and poor credit history.

Request

{
  "Age": 18,
  "Income": 10000,
  "LoanAmount": 1000,
  "CreditScore": 300,
  "MonthsEmployed": 0,
  "NumCreditLines": 1,
  "InterestRate": 2,
  "LoanTerm": 12,
  "DTIRatio": 0.1,
  "Education": "Bachelor's",
  "EmploymentType": "Full-time",
  "MaritalStatus": "Divorced",
  "LoanPurpose": "Other",
  "HasMortgage": "Yes",
  "HasDependents": "Yes",
  "HasCoSigner": "Yes"
}

Response

{
  "Default": 1,
  "decision": "will_default",
  "probability": 0.512
}

This demonstrates a borderline risk case, where the model is uncertain but leans toward default.

---

🔧 cURL Example

curl -X POST "http://localhost:8000/predict" \
-H "Content-Type: application/json" \
-d '{
  "Age": 30,
  "Income": 45000,
  "LoanAmount": 8000,
  "CreditScore": 680,
  "MonthsEmployed": 36,
  "NumCreditLines": 2,
  "InterestRate": 10,
  "LoanTerm": 24,
  "DTIRatio": 0.3,
  "Education": "Master's",
  "EmploymentType": "Full-time",
  "MaritalStatus": "Single",
  "LoanPurpose": "Business",
  "HasMortgage": "No",
  "HasDependents": "No",
  "HasCoSigner": "Yes"
}'

---

🧩 Tech Stack

• Python
• scikit-learn
• FastAPI
• Pydantic
• Docker
• joblib

---

🎯 Project Highlights

• End-to-end ML system
• Strict input validation
• Probability-based predictions
• Dockerized for reproducibility
• Clear API contract
• Production-oriented logging

---

📌 Portfolio Note

This project focuses on ML inference, system design, and deployment, not just model training. It demonstrates how machine learning models are actually used in production systems.

---

👤 Author

Satadru Halder

---