Encoding Categorical Data

Categorical Encoding

Machine learning requires numerical features.

One-Hot Encoding

pd.get_dummies(df['col']) creates binary columns. OneHotEncoder from sklearn does same.

Drop first category to avoid multicollinearity: drop='first'. Handle unknown categories: handle_unknown='ignore'.

Label Encoding

LabelEncoder encodes categories as integers. OrdinalEncoder for multiple columns.

Useful for tree-based methods that handle integers.

Target Encoding

Mean of target for each category. Requires careful handling to prevent leakage.

Smoothing blends with global mean: (n * category_mean + m * global_mean) / (n + m).

Key Takeaways

1. One-hot creates binary columns for categories
2. Label encoding converts to integers
3. Target encoding requires careful implementation