Определение глубины данных на многомерных выборках

Abstract

Concepts of depth and depth trimmed regions are considered, their main properties and classification are presented. 5 notions of data depth and results of investigating their correspondence with requirements, imposed on depth and regions, are discussed. One of the latest forms of depth zonoid depth and algorithms for depth computation and constructing regions are outlined. These mathematical tools for nonparametric data analysis can be applied in location and dispersion estimation, cluster analysis and risk estimation.