Credit Card Approval

Summary

Description: Predict credit card approvals and risk scores

Tags: credit, finance, risk

Language: Python

Framework: scikit-learn

Model Type: Gradient boosting classifier

Overview

Given information about a customer from their credit card application, the model returns whether or not the customer is approved for a credit card and a credit risk score.

A high risk level is associated with past customers who were 60 days or more overdue on their payments.

Inputs

The input parameters include details about age, occupation, family size, housing, marital status, and education level.

Outputs

The model returns an approved value equal to 1 if the credit card is approved, and a value equal to 0 if the credit card is denied.

The model returns a credit risk_score with a value between 0 and 1. A higher score indicates that the customer presents a high credit risk. A lower score indicates that the customer presents a low credit risk.

Model versions

Model version A was trained using a gradient boosting classifier in scikit-learn.

Model version B was trained using a random forest classifier in scikit-learn.

Feature importance

A plot of feature importance for model version A (gradient boosting classifier) is shown below:

A plot of feature importance for model version B (random forest classifier) is shown below:

Example queries

Approved credit card

Input:

{
  "high_balance": 0,
  "owns_home": 1,
  "child_one": 0,
  "child_two_plus": 0,
  "has_work_phone": 0,
  "age_high": 0,
  "age_highest": 1,
  "age_low": 0,
  "age_lowest": 0,
  "employment_duration_high": 0,
  "employment_duration_highest": 0,
  "employment_duration_low": 0,
  "employment_duration_medium": 0,
  "occupation_hightech": 0,
  "occupation_office": 1,
  "family_size_one": 1,
  "family_size_three_plus": 0,
  "housing_coop_apartment": 0,
  "housing_municipal_apartment": 0,
  "housing_office_apartment": 0,
  "housing_rented_apartment": 0,
  "housing_with_parents": 0,
  "education_higher_education": 0,
  "education_incomplete_higher": 0,
  "education_lower_secondary": 0,
  "marital_civil_marriage": 0,
  "marital_separated": 0,
  "marital_single_not_married": 1,
  "marital_widow": 0
}

Output:

{
  "approved": 1,
  "risk_score": 0.08
}

Denied credit card

Input:

{
  "high_balance": 0,
  "owns_home": 1,
  "child_one": 0,
  "child_two_plus": 0,
  "has_work_phone": 0,
  "age_high": 0,
  "age_highest": 0,
  "age_low": 0,
  "age_lowest": 0.25,
  "employment_duration_high": 0,
  "employment_duration_highest": 0,
  "employment_duration_low": 0,
  "employment_duration_medium": 0,
  "occupation_hightech": 0,
  "occupation_office": 0,
  "family_size_one": 1,
  "family_size_three_plus": 0,
  "housing_coop_apartment": 0,
  "housing_municipal_apartment": 0,
  "housing_office_apartment": 0,
  "housing_rented_apartment": 0,
  "housing_with_parents": 0,
  "education_higher_education": 1,
  "education_incomplete_higher": 0,
  "education_lower_secondary": 0,
  "marital_civil_marriage": 0,
  "marital_separated": 0,
  "marital_single_not_married": 0,
  "marital_widow": 0
}

Output:

{
  "approved": 0,
  "risk_score": 0.77
}

Usage

Input parameters

Output parameters

Preparatory work

Prepare data

Downloaded data files from https://www.kaggle.com/rikdifos/credit-card-approval-prediction:

• application_record.csv
• credit_record.csv

and place them in a hosted data collection, data://algorithmiahq/credit_card_approval/ in this case.

The prepare.py script was then used to prepare the data.

Train model

The train-a.ipynb and train-b.ipynb notebooks were used to train a gradient boosting classifier and a random forest classifier, respectively, using scikit-learn.