MULTI-DIMENSIONAL WEIGHT CLUSTERING METHODOLOGY

Abstract

The combined approach to the analysis of the socio-economic sphere is based on the use of information-mathematical methods. This allows you to expand the possibilities of a monothetic approach, for example, when structuring objects of research or indicators describing the studied objects. This approach is currently part of business intelligence. A multidimensional approach is applied in the classification of multidimensional data. Classical classification methods are complemented by the introduction of latent factors. The classification procedure based on latent factors without weight factors is described. In addition to expert methods, a methodology for estimating weight coefficients in multidimensional clustering is proposed. Given the structure of latent integrated indicators, the methodology of formation of weighting factors is considered. The paper proposes structuring using non-metric multidimensional scaling techniques, including taking into account weighting factors. Given the ranking positions of clustering objects, a grouping is carried out in the theoretical space of stimuli. The share ratio of indicators both in latent factors and in the theoretical scale space allows us to exclude the use of expert estimates and their subjectivity. The paper also notes the possibility of combining and dynamic analysis in the analysis of multidimensional data arrays. As an object of study, indicators of the non-production sphere of the Volga Federal District are considered.