A new method of Multi Dimensional Scaling

Abstract

This paper presents a new algorithm called “Population” that is an efficient and high speed method of performing Multi Dimensional Scaling based only on the calculation of a local fitness. Population is not bound to a specific Cost Function but is possible to define its in relation to the considered objective. The motivation for its creation was for use in the elaboration of datasets of great dimensions. In performance comparisons between Population and the Sammon method, Population has consistently excelled. Because of the nature of the algorithm, it is not necessary for the data set to be complete at the moment of the elaboration, for new data can be introduced dynamically in the system.