COMPEL Glossary / cross-validation
Cross-Validation
Cross-validation is a statistical technique for evaluating AI model performance by partitioning data into multiple subsets, systematically training the model on some subsets while testing on others, and averaging the results across all partitions.
What this means in practice
The most common form is K-fold cross-validation, where data is split into K equal parts. Cross-validation provides more reliable performance estimates than a single train-test split because it evaluates the model across multiple data samples, reducing the risk that a lucky or unlucky data split produces misleading results. For enterprise AI governance, cross-validation results provide stronger evidence of model reliability than single-split evaluations, which is important for regulatory compliance and audit preparedness. The COMPEL Evaluate stage recommends cross-validation as standard practice for model performance assessment.
Why it matters
A single train-test data split can produce misleading performance estimates if the split happens to be lucky or unlucky. Cross-validation provides more reliable performance evidence by evaluating models across multiple data partitions, reducing the risk of overconfident deployment decisions. For regulatory compliance and audit preparedness, cross-validation results provide stronger evidence of model reliability than single-split evaluations.
How COMPEL uses it
Cross-validation is recommended as standard practice during the Evaluate stage for model performance assessment. The Technology pillar assesses cross-validation capability during Calibrate as a maturity indicator for model evaluation practices. During Model, evaluation standards including cross-validation requirements are codified. The Produce stage implements cross-validation in ML pipelines, and results are documented as governance evidence for audit purposes.
Related Terms
Other glossary terms mentioned in this entry's definition and context.