ATPA: Explain a preprocessing choice through its consequences
A missing-value method, transformation or category grouping changes the model’s information. Justifying a choice requires more than saying it improved accuracy. Consider what missingness means and whether the same transformation can be applied later.
Worked example or practice scenario
For a skewed positive variable, a log transformation may reduce the influence of very large values. Zeroes require explicit handling; negative values may signal a different variable definition or a data issue. Adding a constant changes interpretation and should be documented.
Try this next
Use a practice dataset to compare summaries before and after transformation. Check out-of-sample behaviour and interpret the resulting coefficients or predictions on the correct scale. Distinguish a genuine extreme observation from an impossible value instead of deleting both automatically.
Reading sources
ActNet editorial guide · October 1, 2026 · Original illustrative examples.