Credit ML & compliance, defined
Plain-English definitions of the terms that come up in credit modeling, fraud ML, fair-lending compliance, and model risk management. Written for risk officers, ML engineers, and analysts working in regulated finance.
Model risk & compliance
Regulatory frameworks and disclosures that shape how credit models are built, validated, and deployed.
SR 11-7 (Model Risk Management)
Joint Federal Reserve / OCC supervisory guidance defining how US banks must govern model risk across development, validation, and use.
Adverse action codes
ECOA-required disclosures explaining to a denied applicant the principal reasons their credit application was rejected.
Disparate impact testing
Statistical analysis evaluating whether a credit policy or model produces disproportionately adverse outcomes for a protected class.
Model performance
Statistics for measuring discrimination, calibration, and distribution stability.
Kolmogorov-Smirnov (KS) statistic
The maximum vertical distance between two cumulative distribution functions — in credit, between the score CDFs of good and bad borrowers.
Gini coefficient
A summary measure of model discrimination, equal to 2 × AUC − 1 in a binary classification setting.
Population stability index
A statistic measuring how much a population’s distribution has shifted between two time periods or samples.
Modeling techniques
Methods used inside the model-building loop that affect performance and interpretability.
Reject inference
A statistical technique used to estimate the credit behavior of rejected applicants who have no observed performance outcome.
Feature importance vs. SHAP values
Two complementary approaches to attributing a model’s output to its input features — global vs. local, marginal vs. additive.