Conference
French
ID: <
10670/1.vmsj5i>
Abstract
National hearing. We are interested here in a particular type of complexity, which is that of links between variables. For example, in economics, household poverty is linked in a complex way to the socio-economic situation of both spouses before their marriage. The first source of complexity comes from the variability of individuals, which calls for the use of statistical modelling (regression models, ANOVA, etc.) to identify and validate these links. But a second source of complexity, which makes this classic statistical framework ineffective, is the large number of variables and the difficulty of assigning them a well-defined type (variables with missing, qualitative or ordinary values, etc.), which calls for more flexible learning methods. We present such a methodology, based on reasons, and illustrate it by applying it to data from a questionnaire on Tunisian households.