AI & Data Quiz

Answers and explanations appear as you take the quiz.

1. A model performs poorly despite having many records. Which guide principle helps explain why?
   • More examples always remove measurement errors
   • The most common label guarantees reliable predictions
   • Data quality, relevance, and coverage matter for the task
   • Training data size alone determines model usefulness
2. A supervised training table pairs each input with the outcome it should predict. Which field acts as the label?
   • The target outcome associated with an example
   • The computer file name
   • A feature available before the prediction
   • A privacy permission
3. A team predicts cancellations at the start of June. Why should it exclude a cancellation reason recorded after the customer leaves?
   • Because smaller datasets always make more accurate models
   • It is a feature not available at the prediction time and leaks the outcome
   • It is a label that must be used as an input
   • It changes the customer’s example unit
4. Why keep multiple photos of the same object together when splitting a recognition dataset?
   • To make the image labels identical
   • To increase the number of independent objects
   • To remove all test examples
   • To prevent near-duplicate views from leaking across train and test
5. Why can replacing every missing value with zero be misleading?
   • Zero may mean a real quantity while missing means unknown
   • Zero is always the correct estimate
   • Missing values cannot carry information
   • Models ignore zeros automatically
6. A sensor value falls outside its plausible physical range. Which response follows the guide’s data-quality checks?
   • Whether the file has a short name
   • Review the measurement and flag a suspicious value rather than assuming it is valid
   • Whether the reading was collected on a weekend
   • Whether more records can be added
7. What does a public data URL establish about reuse?
   • That the dataset is free to use for every purpose
   • That every record is accurate
   • Only that the resource can be accessed publicly, not that all reuse is permitted
   • That consent and retention rules do not apply
8. Why can a label be an imperfect proxy?
   • It may measure a recorded process rather than the real-world outcome of interest
   • Labels always match outcomes
   • A proxy eliminates selection bias
   • Training data do not contain labels