ENSEMBLE METHODS FOR RANKING DATA

Abstract

The last years have seen a remarkable flowering of works about the use of decision trees for ranking data. As a matter of fact, decision trees are useful and intuitive, but they are very unstable: small perturbations bring big changes. This is the reason why it could be necessary to use more stable procedures, as ensemble methods, in order to find which predictors are able to explain the preference structure. In this work ensemble methods as BAGGING and Random Forest are proposed, from both a theoretical and computational point of view, for deriving classification trees when ranking data are observed. The advantages of these procedures are shown through an example on the SUSHI data set.