Grouping of objects in a space of heterogeneous variables with the use of taxonomic decision trees
Vladimir Berikov · Pattern Recognition and Image Analysis · 2011
A problem of classification of objects in the presence of heterogeneous (qualitative, ordinal, nominal, and Boolean) variables is considered. Taxonomic decision trees are used to solve the problem. A quality criterion for a tree is introduced that is based on the Bayesian estimate of the Kullback-Leibler distance between distributions. Statistical modeling is applied to show the efficiency of an algorithm for constructing a tree that uses this criterion.