Data Structure Influence on Mapping Error

Abstract

Various procedures are used for mapping of multidimensional data onto the plane. Sammon’s method of simultaneous nonlinear mapping onto the plane is very popular. In addition, sequential nonlinear mapping has been created for watching the data in real time. The essence of mapping is to preserve the inner structure of distances among the vectors in multidimensional space after mapping them onto the plane. The mapping error characterizes the mapping quality, and it depends on initial conditions and data structure. The paper deals with investigations of how data structure influences on mapping quality. Plenty experiments have been executed at various number of parameters and various number of vectors using regular data (having information about clusters) and random data. The experiments allowed to draw conclusions that mapping error increases increasing the number of parameters, and it remains almost constant increasing the number of vectors. Besides, mapping error of regular data is several times less than that of random data. Ill. 11, bibl. 7 (in Lithuanian; summaries in Lithuanian, English and Russian).