Using genetic programming for the induction of oblique decision trees

Abstract

In this paper, we present a genetically induced oblique decision tree algorithm. In traditional decision tree, each internal node has a testing criterion involving a single attribute. Oblique decision tree allows testing criterion to consist of more than one attribute. Here we use genetic programming to evolve and find an optimal testing criterion in each internal node for the set of samples at that node. This testing criterion is the characteristic function of a relation over existing attributes. We present the algorithm for construction of the oblique decision tree. We also compare the results of our proposed oblique decision tree with the one of C4.5 algorithm.