PCA

PCA (Principal Component Analysis) reduces the number of dimensions in data by finding the orthogonal axes (principal components) along which the data varies most, then projecting onto the top few. It compresses correlated features into fewer, uncorrelated ones while preserving most of the variance.

Worked example: 100 correlated sensor readings might be summarized by their top 3 principal components that capture 95% of the variance — turning a 100-dimensional problem into a 3-dimensional one you can plot and model. Gotcha: PCA is linear (it only captures linear structure) and its components are combinations of all original features, so they are hard to interpret — for nonlinear structure or visualization, methods like t-SNE and UMAP are used, though those distort global distances and are for viewing, not as model inputs.