Multi-Class Budget Exploratory Trees

Abstract. This paper provides a new method to grow exploratory classification trees in multi-class problems. A two-stage algorithm, using recursively the latent budget model, is proposed to find ever finer partitions of objects into prior fixed number of groups. A new rule to assign the class labels to the children nodes is considered to deal with fuzzy data. Then, a software prototype namely E.T. Exploratory Trees, developed in the Matlab environment, is proposed to show the main features of the methodology through several interactive graphic tools.