Credit risk scorecard using historical repayment data

Project Description

Four years of lending data, roughly 18,000 loans, with repayment outcomes. We want a scorecard we can defend to a regulator and explain to a borrower — which rules out anything we cannot interpret. A gradient boosted model that scores better and cannot be explained is not a better model for us. We need the usual rigour: out-of-time validation, stability over cohorts, and clear documentation of every variable and why it is in there. Please describe a scorecard you have built and how it performed after deployment, not just in validation.

Skills Required