🏷️ 03 — Categorical Features

Machine learning models need categories converted to numbers.

---

One-Hot Encoding

from sklearn.preprocessing import OneHotEncoder

encoder = OneHotEncoder(handle_unknown="ignore")

Good for nominal categories:

• city
• product type
• browser

---

Ordinal Encoding

Use when categories have real order.

mapping = {"low": 1, "medium": 2, "high": 3}
df["risk_level_num"] = df["risk_level"].map(mapping)

Do not ordinal-encode unordered categories like city.

---

Rare Categories

High-cardinality categories can overfit.

top = df["city"].value_counts().head(10).index
df["city_clean"] = df["city"].where(df["city"].isin(top), "Other")

---

Target Encoding Warning

Target encoding can be powerful but dangerous because it can leak target information.

Use cross-validation-based target encoding if needed.

---

Next

➡️ 04-datetime-and-text-features