Preserve the evaluation population
Training interventions can help a learner attend to rare outcomes, but evaluation must reflect the decision population. Split first, then resample training data only. Otherwise duplicated examples can cross partitions and inflate results. Keep the natural class prevalence in the holdout unless the evaluation design explicitly accounts for sampling.
A dataset contains 100 positive and 9,900 negative cases. Explain why oversampling before splitting leaks duplicate examples.
Distinguish weights from duplication
Class weights increase the loss contribution of selected examples. Oversampling changes how often examples appear during training. Neither creates new independent evidence about rare cases. Compare these interventions with a baseline using fixed splits, and inspect mislabeled positives: emphasizing a noisy example can magnify its harm.
Give a reason why repeating ten positive examples 100 times does not provide the information of 1,000 distinct positives.
Recheck probabilities and thresholds
A model trained with a changed class balance may produce scores that are not calibrated for the deployment population. Assess calibration on representative validation data and select an operating threshold against review capacity and error costs. Do not assume a score of 0.5 is the appropriate release threshold after weighting.
A review team can handle 200 cases daily. Describe how you would choose and validate an operating threshold.
Report the tradeoff
Suppose a candidate finds 80 of 100 positives but flags 400 negatives. Recall is 80%, while precision is 80/480, about 16.7%. Higher recall may still create an unaffordable queue. Report both errors and operational load, and test relevant slices before choosing the intervention.
If capacity is 300 reviews, explain why this 480-alert candidate fails even with strong recall.
Where should oversampling happen?
Keep exploring
Use the official certification page and its linked resources to validate your coverage. For guided labs, explore the Google Skills learning path. Revisit the practical exercise and explain your decision in your own words before moving on.