“On Lines and Planes of Closest Fit to Systems of Points in Space” by Karl Pearson was published in the Philosophical Magazine in November 1901. It is generally credited as the origin of principal component analysis (PCA), the most widely used method for reducing the dimensionality of data.
Pearson posed a geometric question: given a cloud of points, what line or plane best fits them in the sense of minimizing the total squared perpendicular distance from the points to it? His answer identified the directions of greatest variance in the data. Those directions, later called principal components, form a new set of axes in which the first axis captures as much of the spread as possible, the second captures the most of what remains, and so on. Projecting data onto the first few components compresses it while keeping most of its structure.
PCA predates computers by decades but became a staple of statistics and machine learning once they arrived. It is used to visualize high-dimensional data, remove redundant features, denoise signals, and speed up downstream models. Newer nonlinear methods such as t-SNE and UMAP address cases where PCA’s straight-line assumptions are too limiting, but PCA remains the first tool most analysts reach for.
Why business readers should care: PCA is the quiet engine behind much of data compression, anomaly detection, and the “reduce 500 columns to 10 useful ones” step that makes large datasets tractable.