ABSTRACT
Traditional multivariate statistics has approached variable selection using stepwise selection and best subset selection within linear-regression models. More recent trends are nonlinear models and addressing the question of instability (a small change in the data might result in a drastic change in the inferred results). This chapter discusses an approach that covers both of these concerns. ¡Nonlinearity is addressed using decision trees as the underlying regressors or classifiers, and instability is addressed by employing ensembles of decision trees.
