An Application of a Nonmetric Multidimensional Scaling Method for the Reduction of Dimensionality in the Problems of Pattern Recognition

摘要

A space called joint space is a Euclidean vector space in which each pattern and each feature representative is described by separate vectors. The technique employed in this study for transforming data representatives into a common joint space belongs to the nonlinear mapping algorithms which are known as the Guttman-Lingoes multidimensional scaling methods. The acceptable configuration of data vector representatives in the corresponding joint space must reflect with reasonable accuracy the geometrical relationships existing within the data configuration before the application of a particular multidimensional scaling method. In general, the Guttman-Lingoes multidimensional scaling methods could be applied to both binary and continuous data, in any kind of pattern recognition problem, and for any number of pattern and feature representatives. (Author Modified Abstract)