Skip to content

Topic Drill 5.1.1: Input Data Risks and Mitigations ​

About this drill

LO AI-5.1.1 (K2): Give examples of test approaches used for the risk mitigation of input data for a machine learning system.
Lesson: Ch 5, section 5.1.1

This drill covers every point in section 5.1 / 5.1.1:

  • The objective and contents of input data testing
  • Every row of the risks and mitigations table: bias, provenance, poisoning (four mitigations), constraints, features, representativeness (five risks), labels, pipeline (four risks)

Topic Drill: 5.1.1

Q1. A road-scene dataset for a driver-assistance model contains 95% daytime images, although the product will also be used at night. Which test approach from the 5.1.1 table addresses this risk?

ADataset constraint testing
BData representativeness testing
CLabel correctness testing
DData provenance testing

Q2. Which TWO of the following does the syllabus list as mitigations for POISONED training data?

Select 2 options.

ALabel correctness testing
BFeature testing
CA/B testing
DDataset constraint testing
EExploratory data analysis (EDA)

Q3. A numeric "annual_income" column contains entries such as "N/A" and "unknown". Which mitigation fits this risk BEST?

ATesting for bias
BData provenance testing
CFeature testing
DDataset constraint testing

Q4. A model is trained on 300 input attributes, many of them redundant, and the team suspects the wrong attributes were chosen. Which mitigation does the syllabus map to "sub-optimal feature selection"?

AFeature testing
BData pipeline testing
CLabel correctness testing
DA/B testing

Q5. Annotators labelled images of house cats and lynxes inconsistently, because the guidelines did not say how to handle blurry photos. Which mitigation applies?

AData provenance testing
BData representativeness testing
CLabel correctness testing
DData pipeline testing

Q6. After a schema change, the nightly job that transforms raw sensor data for the model silently drops some records. Which mitigation applies?

ATesting for bias
BData pipeline testing
CFeature testing
DLabel correctness testing

Q7. To boost a rare class, the team generated thousands of synthetic examples of it, and now that class dominates the dataset. Which risk is this, and which mitigation applies?

APoisoned training data → red teaming
BWrong data types → dataset constraint testing
CPoor annotation → label correctness testing
DSkewed dataset through data augmentation → data representativeness testing

Q8. Which mitigation does the syllabus map to "problems within the algorithm, model, or ML development framework that introduce systemic unfairness"?

ATesting for bias
BData provenance testing
CData pipeline testing
DFeature testing

Q9. Training data is stored on shared drives with no versioning, no access control and no record of who changed what. Which mitigation fits "data is poorly managed"?

ADataset constraint testing
BLabel correctness testing
CData provenance testing
DFeature testing

Q10. What is the objective of input data testing, according to section 5.1?

ATo check that the model meets its accuracy target
BTo confirm that the data used by the ML system for training, testing and prediction is of sufficient quality
CTo verify the user interface of the ML system
DTo compare two versions of the model in production

0 / 10 answered

← 4.3.2 Risk-Based Testing · Next drill: 5.1.2 Testing for Bias →

Unofficial study guide for the ISTQB® Certified Tester AI Testing (CT-AI) exam. Not affiliated with International Software Testing Qualifications Board (ISTQB®).