Loan Delinquency Prediction
Prediction of charge-off and customer delinquency for a micro-finance company
Problem Statement
A micro-finance company that gives small loans to businesses in the US and Canada was facing high charge-off rates. A predictive modeling framework was required for early identification of high risk accounts.
Challenges
- No dedicated analytics team in the company, so no prior work to build on
- Data spread out across multiple unclean datasets
- Lack of a standard definition of charge-off
Solution
Summary
I developed a supervised machine learning solution to estimate the probability of account delinquency.
Approach
In addition to data cleaning and transformation, the approach involved:
- Identifying key hypotheses to detect charge-off risks
- Defining charge-off accounts, with special consideration for accounts with no payments
- Building a data processing and metrics computation pipeline
- Conducting exploratory analysis and feature selection
- Developing and evaluating multiple classification models
- Creating a prioritization framework to optimize recovery efforts
Deliverable
After testing various classification models, the most effective one was deployed on-site for periodic predictions. Key predictors of delinquency were identified, and actionable recommendations were provided to help prevent future charge-offs.
Modeling Approach
Modeling Results
Account Prioritization Framework
Results & Impact
Model Accuracy
Model Recall
Est. Value Benefit to Company
I am truly happy to have found and contracted Nanda. He is very knowledgeable and did a remarkable job on the project.
Marco M.
Analytics Manager, Dublin
Have a similar challenge?
Let's discuss how we can develop solutions for your specific use case.