[ML] Credit Card Default Prediction

Objective

• Predict potential credit card defaulters using customer demographic and transaction data.

• Develop a model to proactively manage high-risk customers and improve financial liquidity.

  
  

Dataset Overview

• Source: [UCI Machine Learning Repository]

• Description: Binary classification of default occurrences (1: default, 0: non-default).

• Total Records: 26,457

• Number of Features: 20

  
  

Problem Definition

• Challenge: Rising defaults due to aggressive marketing increasing new card issuances.

• Risk: Delayed recovery of credit card dues leading to liquidity issues.

  
  

Expected Impact

• Increase in financial liquidity through efficient recovery from defaulters.

• Leverage the default prediction model for customer management programs and marketing campaigns.

  
  

Approach

1. Data Collection and Preparation

   • Collect and preprocess customer data, including credit card usage history and demographic information.

2. Model Development

   • Utilize recent transaction history and customer features to train the prediction model.

   • Focus on binary classification to identify high-risk customers.

3. Model Deployment and Operations

   • Implement the model for monthly extraction of high-risk customers.

   • Use predictions to manage defaulters and run targeted marketing campaigns.

     
     

Key Code Highlights

• Handle missing values (df['occyp_type'].fillna('etc')).

• Process credit labels and transform age bins (optb.transform(df['DAYS_BIRTH'], metric="bins")).

  
  

Practical Application

• Use model predictions to enhance liquidity by managing defaulters effectively.

• Increase revenue through targeted marketing for high-risk customers (e.g., additional products, revolving credit).