How to make sammon's mapping useful for multidimensional data structures analysis

摘要

The ways of optimization of Sammon's mapping technique are suggested. Two sorts of the mapping artifacts produced by the local minimum traps and non-coherence of the source and target spaces (i.e. multi- and low-dimensional ones, respectively) are discussed. The methods of reduction of the artifacts' influence on the resulting two- and three-dimensional patterns are proposed. The nuclear reactor diagnostics system is taken as the source of real multidimensional data. Usefulness of the mapping for their analysis is shown.