Feature selection vs. feature extraction

Feature selection keeps a subset of the original features and discards the rest. The result is interpretable — you still know what each feature means — and you don't need to measure discarded features in future.

Feature extraction builds new features as combinations of the originals. PCA creates linear combinations; autoencoders, t-SNE and UMAP create non-linear ones. It can compress information more efficiently, but the new features are harder to interpret.

• Use selection when interpretability or measurement cost matters, or many features are irrelevant.
• Use extraction when features are many and correlated (pixels, sensor readings), or you need maximum compression or a 2-D picture.