Credit scoring
Predict borrower outcomes with calibrated, time-aware models
Credit scoring predicts repayment or default outcomes from borrower information. Application scores use information available at origination; behavioural scores update with subsequent account history. Statistical and machine-learning models can be used, but the prediction horizon, default definition and deployment population must be specified. Discrimination, probability calibration and decision consequences are separate assessment tasks.
Choose this for predicting a clearly defined borrower outcome when appropriately governed historical predictors and later repayment outcomes are available. Evaluation must reflect the future applicant population and real information timing.
Strengths
- Provides reproducible and scalable predictive assessment
- Supports separate examination of ranking and calibrated default probability
Limitations
- Rejected applicants often lack observed repayment outcomes
- Changing populations, imbalance and unfair proxy effects can impair validity
Know the boundary
A score’s predictive accuracy does not establish that its predictors cause default or justify an unexamined lending rule.