The process of reducing the dimensionality (i.e. the number of variables) of multivariate data by deriving a small number of new variables that contain much of the information in the original data.