Landscape estimation of decision-tree induction based on grammatical Evolution using rank correlation

The type of evolutionary machine learning known as grammatical Evolution (GE) is currently receiving a great deal of attention. GE is particularly suitable for developing decision-tree classifiers because of a framework, in which candidate solutions are generated via production rules. Various decision-tree classifier methods based on GE have been proposed. In general, the performance of GE systems is improved by enhancing the genetic diversity of the candidate solutions. Therefore, most GE methods are focused on the initialization of solutions. However, it is known that an effective search bias based on a landscape is also essential for evolutionary computation methods. Unfortunately, because of their solution structures, GE-based decision-tree classifiers can not form a unique landscape in terms of an objective function as can real-valued optimization problems. In this paper, we present a method for estimating a landscape using rank correlation based on two types of features extracted from GE solutions, and we apply it to well-known benchmark problems. We show that the proposed method can capture a landscape effectively. To the best of the authors' knowledge, this is the first study to report about a landscape estimation method based on GE solutions. The results in this paper help with understanding how to establish suitable a search bias for GE-based decision-tree classifiers.