Multiple correspondence analysis of a crosstabulations matrix using the Kohonen algorithm

摘要

The multiple correspondence analysis is a statistical technique to handle qualitative variables and to try to show the correlations between several kinds of variables in a population sample. Classical methods like canonical analysis or factorial analysis are proven to be efficient to deal with this sort of problems. But they present some inconvenients: they are intrinsically linear and moreover they provide graphic representations wich have no true significance overall when there are more than two crossed variables. In an previous paper [3], M.Cottrell et al. had defined a nwe algorithm (KOUPLET) wich allows to qualitative variables. This algorithm is inspired from the self organisation Kohonen algorithm. In this paper, we present another Kohonen-like algorithm to analyze the relations between Q qualitative variables Q>2.