Pattern Recognition, 4th Edition

15.10.1. Dimensionality Reduction Clustering Approach

The general idea of this approach is to identify an l′-dimensional space H^l′ (l′ < l), project the data points in X onto it, and apply a clustering algorithm on the projections of the points of X into H^l′. For the identification of H^l′ one may use (a) feature generation methods, (b) feature selection methods, and (c) random projections. In the sequel, we “touch” briefly the first two methodologies, in that they build upon the techniques treated in more detail in Chapters 5 and 6, and continue to pursue the random projection method.

Feature generation methods, such as the principal component analysis (PCA) and the singular value decomposition (SVD), generally preserve the distances between ...