Hands-On Exercise: Metamorphic Testing CT-AI Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard

   A team is testing an AI system that predicts stock prices. They define a metamorphic relation: if the input features are scaled by a factor of 10 (e.g., all prices and volumes multiplied by 10), the predicted price should also be scaled by a factor of 10. They run a source test case and get a prediction of $100. They create a follow-up test case by scaling all inputs by 10. The follow-up prediction is $1,000. However, the team is concerned that this relation might not be valid for the system. What is the most likely reason for this concern?

   • AThe relation is not valid because scaling inputs by 10 is not a realistic transformation for stock price prediction.
   • BThe relation assumes the model is linear, but many AI models are non-linear and may not scale outputs proportionally.
   • CThe relation is too weak because it only checks for proportional scaling, not exact equality.
   • DThe relation is not applicable because stock price prediction is a regression task, not a classification task.
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A team is testing an AI system that predicts house prices. They define a metamorphic relation: adding a bedroom should increase the predicted price. They have a source test case with a house that has 3 bedrooms and a predicted price of $300,000. They create a follow-up test case with 4 bedrooms. What is the next step in executing the metamorphic test?

   • ARun the follow-up test case through the AI system and record the predicted price.
   • BManually calculate the expected price for the 4-bedroom house and compare it to the source output.
   • CModify the source test case to have 4 bedrooms and run it again, discarding the original source output.
   • DSkip the execution and directly check if the relation holds based on the source output alone.
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   Why is metamorphic testing particularly useful for testing AI-based systems?

   • AIt eliminates the need for any test data.
   • BIt provides a way to verify outputs when a test oracle is unavailable or impractical.
   • CIt guarantees that the AI system is free of defects.
   • DIt automatically generates the expected output for each test case.
   Reveal Solution Discussion Select an answer first
5. 9foundation · easy

   What does a failed metamorphic test indicate?

   • AA potential defect in the AI system that needs investigation.
   • BThat the AI system is working correctly.
   • CThat the metamorphic relation is definitely incorrect.
   • DThat the test data is invalid.
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A team is testing an AI system that detects fraudulent transactions. They define a metamorphic relation: if the transaction amount is increased by a small amount (e.g., $1), the fraud classification should not change. They run a source test case with an amount of $100 (output: not fraud) and a follow-up test case with an amount of $101 (output: fraud). The team is surprised by this result. What is the most appropriate next step?

   • AIgnore the result because the relation is not valid for fraud detection systems.
   • BInvestigate the model's behavior to understand why a $1 increase caused a change in classification, as this could indicate a defect or a legitimate boundary condition.
   • CConclude that the model is working correctly because fraud detection is inherently sensitive to small changes.
   • DModify the relation to allow for changes in classification when the amount changes.
   Reveal Solution Discussion Select an answer first