Production

Credit Scoring & Risk Model

Biznesni Rivojlantirish Banki

⚡Credit approvals depended on inconsistent manual assessment - similar applicants could get different outcomes, with no quantified basis for the decision.

⚠️Default risk was hard to forecast without a model that learned from the bank's own repayment history.

⚙️Built classification models estimating probability of default from income, transaction behaviour, existing obligations, and account tenure.

🛡️Calibrated the model so scores map to meaningful risk bands, and validated against historical defaults to confirm discrimination held on unseen applicants.

🚀Delivered repayment-ability scoring that gives credit officers a consistent, defensible, data-driven signal for every application.

↳ ML credit-scoring models assessing borrower repayment ability - data-driven, auditable lending decisions in production at a commercial bank

Credit RiskClassificationScikit-learnPyTorchModel CalibrationPython