What is Model Evaluation

Definition

Model evaluation is the systematic assessment of how well a machine-learning model performs, using metrics, benchmarks, and test datasets to measure accuracy, robustness, fairness, and fitness for purpose before and after deployment.

« Back to Glossary Index

Quantifies model quality against clear metrics before deployment
Compares candidate models objectively to select the best fit
Surfaces weaknesses such as bias, overfitting, or poor edge-case handling
Establishes baselines to detect degradation after release

Real World Example

Before launching a fraud model, a team evaluates it on a held-out dataset measuring precision, recall, and false-positive rate across customer segments, catching that it underperforms for new accounts and adjusting before release.

FAQs

What metrics are used in model evaluation?

Common metrics include accuracy, precision, recall, F1, and AUC for classification, and task-specific measures for other models, including fairness and robustness checks.

Why evaluate on a held-out dataset?

Testing on data the model never trained on gives an honest estimate of how it will perform on new, unseen inputs.

How is LLM evaluation different?

LLM evaluation often combines automated metrics with human judgment and benchmark suites, since output quality is harder to score numerically.