A parks department wants to decide which tree-canopy classifier generalizes better on holdout wards by reviewing two training jobs together. What should they do?

Comparing model performance across the two Azure Machine Learning jobs using holdout-ward metrics lets the parks department decide which tree-canopy classifier generalizes better. Side-by-side performance metrics on unseen wards are the evidence Domain 2 expects for job comparison—not training-ward metrics, fairness assessments, or explainability comparisons. Holdout evaluation protects against picking a model that only memorized training wards.

Comparing model performance across the two Azure Machine Learning jobs using training-ward metrics fails the stem's goal: training-ward metrics reflect fit on data the model already saw, not generalization to the holdout wards the parks department actually needs to predict. Comparing model fairness across the two Azure Machine Learning jobs using holdout-ward metrics fails the stem's goal: fairness assesses equity across groups, not which classifier has better overall performance and generalization to unseen wards. Comparing model explainability across the two Azure Machine Learning jobs using holdout-ward metrics fails the stem's goal: explainability shows why a model makes predictions, not which classifier performs better on holdout wards.

Exam caveat: Job-comparison questions require performance-based evaluation on holdout data; training-ward metrics, fairness comparisons, and explainability comparisons are distractors.

Operational check: ensure both jobs logged the same holdout-ward metrics, open the job comparison view or export metrics, select the better generalizing classifier, and retain both metric sets in experiment history for audit.