Loan Default Predictor

Predict whether a loan applicant is likely to default on a loan based on personal, financial, and credit attributes.

Classification

AutoML (Azure Machine Learning)

—

Given applicant data, the model predicts:

• 0 → Will likely default on loan
• 1 → Will likely not default on loan

Azure AutoML trained and selected the best performing algorithm based on metrics like AUC-weighted and accuracy.

—

Used the Azure ML Studio UI to upload a CSV dataset (e.g., `loan_applicants.csv`), containing labeled data with these columns:

["ApplicantID", "Age", "MaritalStatus", "Dependents", "Education",
 "AnnualIncome", "EmploymentType", "JobSector", "CreditScore",
 "LoanAmount", "LoanTermMonths", "LoanPurpose", "InterestRate",
 "PastDefaults", "LatePayments", "BankCustomerYears", "OwnsHome", "LoanDefault"]

Where `LoanDefault` is the target label (0 or 1).

Used the Azure ML Studio UI to:

• Start a new AutoML classification run
• Select `LoanDefault` as the target column
• Choose compute cluster and enable explanation and metrics logging
• Let AutoML train multiple models and select the best one

After training:

• Selected the best run
• Clicked Register Model
• Model is now available in the workspace under Assets > Models

Deployed the model from Azure ML Studio:

• Clicked Deploy > Real-time endpoint
• Selected compute target
• Endpoint URL was generated and secured with an API key

Access keys from Key Vault (Python):

api_key = secret_client.get_secret("max-ml-key").value
url = secret_client.get_secret("max-ml-endpoint").value
 
if not api_key or not url:
    raise Exception("Missing key or URL from Key Vault")

Sample Prediction Request:

import urllib.request
import json
 
data = {
  "input_data": {
    "columns": [
      "Age", "MaritalStatus", "Dependents", "Education", "AnnualIncome",
      "EmploymentType", "JobSector", "CreditScore", "LoanAmount",
      "LoanTermMonths", "LoanPurpose", "InterestRate", "PastDefaults",
      "LatePayments", "BankCustomerYears", "OwnsHome"
    ],
    "index": [0],
    "data": [[35, "Married", 2, "Graduate", 55000, "Full-time", "Private", 710,
              25000, 60, "Home", 5.2, 0, 1, 6, "Yes"]]
  }
}
 
body = json.dumps(data).encode("utf-8")
headers = {
    "Content-Type": "application/json",
    "Authorization": f"Bearer {api_key}"
}
 
req = urllib.request.Request(url, body, headers)
 
try:
    with urllib.request.urlopen(req) as response:
        decoded = response.read().decode("utf-8")
        prediction = json.loads(decoded)[0]
 
        if prediction == 0:
            print("Prediction: Will likely default on loan")
        elif prediction == 1:
            print("Prediction: Will likely not default on loan")
        else:
            print(f"Unexpected prediction value: {prediction}")
 
except urllib.error.HTTPError as error:
    print("HTTP error:", error.code)
    print(error.read().decode())

—

{
  "input_data": {
    "columns": [
      "Age", "MaritalStatus", "Dependents", "Education", "AnnualIncome",
      "EmploymentType", "JobSector", "CreditScore", "LoanAmount",
      "LoanTermMonths", "LoanPurpose", "InterestRate", "PastDefaults",
      "LatePayments", "BankCustomerYears", "OwnsHome"
    ],
    "index": [0],
    "data": [[35, "Married", 2, "Graduate", 55000, "Full-time", "Private", 710,
              25000, 60, "Home", 5.2, 0, 1, 6, "Yes"]]
  }
}

—

The endpoint returns either:

• [0] or [1]

Mapped in your script to readable text:

• 0 → “Will likely default on loan”
• 1 → “Will likely not default on loan”