Density regression trees

Symbolic data analysis was introduced in the last decade in order to provide a generalization of standard data analysis tools. This class of data allows each individual (or class) to take a finite set of values, or categories, intervals or distributions on each variables. In literature this kind of data are known as multi-valued variables. We focus on the first category mentioned above. As a matter of fact, supervised classification and non-parametric regression methods are not yet available. We propose a new non-parametric and non-linear regression approach dealing with quantitative multi-valued response variables and heterogeneous classical and symbolic predictors.