Evolution of Decision Trees

This paper addresses the issue of the induction of orthogonal, oblique and multivariate decision trees. Algorithms proposed by other researchers use heuristic, usually based on the information gain concept, to induce decision trees greedily. These algorithms are often tailored for a given tree type (e.g orthogonal), not being able to induce other types of decision trees. Our work presents an alternative way. We propose to induce a decision trees (without regarding the type) with an unified algorithm based on artificial evolution. Experiments were performed with GALE, our fine-grained parallel Evolutionary Algorithm, and another well-known induction technique on several datasets. Results suggest that Evolutionary Algorithms are competitive and robust for inducing all kinds of decision trees, achieving sometimes better performance than traditional