On the quest for easy-to-understand splitting rules Fernando Berzal a, * , Juan-Carlos Cubero a , Fernando Cuenca b , Marıa J. Martın-Bautista a a Department of Computer Science and Artificial Intelligence, E.T.S. Ingenierıa Inform atica, University of Granada, Granada 18071, Spain b Xfera, Madrid, Spain Abstract Decision trees are probably the most popular and commonly used classification model. They are built recursively following a top-down approach (from general concepts to particular examples) by repeated splits of the training dataset. The chosen splitting criterion may affect the accuracy of the classifier, but not significantly. In fact, none of the proposed splitting criteria in the literature has proved to be universally better than the rest. Although they all yield similar results, their complexity varies significantly, and they are not always suitable for multi-way decision trees. Here we propose two new splitting rules which obtain similar results to other well-known criteria when used to build multi-way decision trees, while their sim- plicity makes them ideal for non-expert users. Ó 2002 Elsevier Science B.V. All rights reserved. Keywords: Supervised learning; Classification; Decision trees; Splitting rules 1. Introduction Decision trees are probably the most popular and commonly used classification model; e.g. see [5,16]. Decision trees are built recursively following a top-down approach (from general concepts to particular examples). That is the reason why the acronym TDIDT, which stands for top-down induction on decision trees, is used to refer to this kind of algorithms. www.elsevier.com/locate/datak Data & Knowledge Engineering 44 (2003) 31–48 * Corresponding author. Tel.: +34-958-242376; fax: +34-958-243317. E-mail addresses: fberzal@decsai.ugr.es (F. Berzal), jc.cubero@decsai.ugr.es (J.-C. Cubero), fernando.cuenca@ xfera.com (F. Cuenca), mbautis@decsai.ugr.es (M.J. Martın-Bautista). 0169-023X/03/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved. PII:S0169-023X(02)00062-9