A simple alternative to principal components analysis. To understand the underlying idea, visualize a data set as being a cloud of points in p-dimensional space. Now imagine shining a light on the cloud so that we see the projections of the points on some lower-dimensional ‘wall’. If there is an obvious pattern that was not visible previously, then, indeed, light has been cast upon the problem! Typically, the patterns searched for are non-overlapping clusters.
Subjects: Probability and Statistics.