Guided Notes
Transform: Scaling & Feature Engineering
Normalization and standardization, storing scaling parameters, and creating new features like GDP per capita.
Why scale features?
Slide 1 / 5GDP values run into trillions while populations run into millions. Many machine learning algorithms — those based on distances or gradients — let the larger-scale feature dominate simply because its numbers are bigger.
Scaling puts features on comparable ranges. Two common methods:
• Normalization (min-max) — rescales to the range 0 to 1. • Standardization (z-score) — rescales to mean 0 and standard deviation 1.
The terms are sometimes used loosely; check which one is meant.