Credit Risk Scoring Model
Reduced false rejection rate by 55% while maintaining the same default rate. Approved ₦4.2B in additional loans in the first quarter.
A Nigerian fintech was rejecting 40% of creditworthy applicants due to an outdated rules-based scoring system. Built a gradient boosting model trained on 2M+ loan records, incorporating alternative data signals including mobile money behaviour.
PythonXGBoostPostgreSQLAirflow