Feature Engineering Basics

Models are only as good as their inputs

Feature engineering is transforming raw data into signals a model can use. It frequently beats swapping in a more complex algorithm.

High-impact techniques

• Encoding categories — one-hot for low-cardinality, target/ordinal for high.
• Scaling — standardize numeric features for distance-based models.
• Interactions — combine features (price per square foot).
• Datetime parts — extract day-of-week, month, is_weekend.
• Aggregations — counts and averages per entity.

Rule

Fit any transformation (scalers, encoders) on the training set only, then apply to test data — otherwise you leak information and overstate performance.