A Statistic Approach to NPA Recovery with Machine Learning

Introduction

Managing Non-Performing Assets (NPAs) is crucial for keeping a bank’s portfolio healthy. NPAs are loans where the borrower is not making payments as scheduled. Finding and predicting which customers might recover from NPA status to become regular, paying customers can greatly improve recovery strategies and overall financial health. In this use case, we developed two machine learning models:

1.Collection Amount Model: Predicts how much money can potentially be collected from customers who are currently NPAs.

2. Upgrade Model: Predicts the likelihood of an NPA customer becoming a regular, clean customer again.

Note: An NPA (Non-Performing Asset) refers to a loan or advance where the principal or interest payment has been overdue for 90 days or more. Managing NPAs effectively is crucial for maintaining financial stability.

Photo by Towfiqu barbhuiya on Unsplash

Technical Approach

• Data Preparation: We started with financial data from 2023–2024, cleaning it up by handling missing values, encoding categorical data, and normalizing continuous variables.

Bivariate Analysis:

• Categorical Variables: We used bar charts to explore relationships and distributions.