**Category: `species`.** It is the only classifier in the table, with three balanced groups of 50 specimens each. The dataset's purpose is to separate these groups, so every measure is best read by species. **Metric: `petal_length`.** I chose it because it separates the species better than the other three measurements (my judgement from the species ranges below, not a separate statistical test). **How `petal_length` varies by species** (from `iris_by_species`): - **Setosa:** average 1.46 cm, median 1.5, range 1.0–1.9. - **Versicolor:** average 4.26 cm, median 4.35, range 3.0–5.1. - **Virginica:** average 5.55 cm, median 5.55, range 4.5–6.9. **What this shows** - **Setosa is fully separable.** Its longest petal (1.9) is well below versicolor's shortest (3.0), so petal length alone identifies it with no overlap. Average petal length is about 2.9 times higher in versicolor and 3.8 times higher in virginica. - **Versicolor and virginica overlap only slightly.** Their ranges share a narrow band from 4.5 to 5.1, while their averages differ by about 1.3 cm. - **Sepal width is a weaker separator.** Its ranges overlap heavily across all three species: setosa 2.3–4.4, versicolor 2.0–3.4, virginica 2.2–3.8. Averages are 3.43, 2.77 and 2.97 cm. - **Petal width behaves much like petal length.** Averages are 0.25, 1.33 and 2.03 cm. For choosing classifier features, petal dimensions, led by `petal_length`, are the strongest starting point. The per-species figures come from one query of `iris_by_species`, rendered in the table above. {"action":"final","content":"placeholder"}