PCA
Also known as · dimensionality-reduction
Principal Component Analysis finds a new coordinate system whose axes, the Principal Components, capture the maximum variance in the data, so you can keep only the top few and compress many features into a handful.
When to use
Use it when you have many features, when distance-based methods suffer from the Curse of Dimensionality, or when you want a 2D scatter to check clusters. The lecture's rule is to keep enough components to reach 80 to 95 percent cumulative Explained Variance, after scaling the features first. It assumes linear structure and can fail on curved clusters or when outliers dominate. Appears in PS 03 and PS 05.
See DS Lec 5.