MLA-C01 Practice Question: A data scientist is working with a dataset…

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Binary encoding

  Why it's wrong here

  Binary encoding converts the 200 countries into roughly eight bit columns, so individual coefficients represent bit positions rather than countries, destroying interpretability. It is tempting because it compresses high-cardinality features, and would be correct for tree ensembles or memory-constrained pipelines where per-category meaning is not required.

• ✗

  Target encoding

  Why it's wrong here

  Target encoding replaces each country with a statistic derived from the target variable, which leaks label information into training and yields coefficients that no longer express a per-country effect. It is tempting for high-cardinality features, and would suit gradient-boosted trees, but not an interpretable linear model.

• ✗

  Label encoding

  Why it's wrong here

  Label encoding assigns arbitrary integers to countries, imposing a false ordinal ranking that linear regression treats as a single numeric slope, so it cannot represent 200 distinct effects. It is tempting because it is compact, and would be correct for tree-based models, which split on integer thresholds.

• ✓

  One-hot encoding with dropping the first category

  Why this is correct

  One-hot encoding creates a binary column per country, and dropping one category avoids perfect multicollinearity — the dummy variable trap — which would make linear regression coefficients unstable. Retaining 199 columns preserves interpretability, unlike target or ordinal encoding, which impose false ordering on nominal country values.

Go deeper

Related to this question