Constructive Induction of Cartesian Product Attributes

Constructive induction is the process of changing the representation of examples by creating new attributes from existing attributes. In classi cation, the goal of constructive induction is to nd a representation that facilitates learning a concept description by a particular learning system. Typically, the new attributes are Boolean or arithmetic combinations of existing attributes and the learning algorithms used are decision trees or rule learners. We describe the construction of new attributes that are the Cartesian product of existing attributes. We consider the e ects of this operator on a Bayesian classi er an a nearest neighbor algorithm.