Data pipeline and representativeness · Question 50 of 60
A skin cancer classifier was trained only with images from a single clinic and is to be used nationwide. Which steps of representativeness testing are appropriate here?
3 answers are correct. Hard
Show answer
Answer
- A Define the target population, for example with domain experts and data from similar systems, and create a reference from it using stratified sampling
- B Compare training data and reference data with exploratory data analysis and look for gaps or concentrations
- E Compare distributions with formal tests such as Chi-squared or Kolmogorov-Smirnov
Explanation
The test runs in three steps: define the target population (operational context, experts, existing systems, benchmarks, stratified sampling as a baseline), analyze data characteristics (exploratory data analysis of training and reference data, gaps and anomalies) and apply statistical techniques. It belongs before training so that no model is built on unrepresentative data.