BibTeX
@MISC{Scott_binarydecision,
author = {K. M. Ho and P. D. Scott and K. M. Ho and P. D. Scott},
title = {Binary Decision Trees},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
. In this paper we consider the possibility of overcoming fragmentation during the construction of decision trees by dividing the values of discriminating nominal attributes into two groups and constructing only two subtrees, one for each group. We discuss some of the problems and limitations that may arise using the conventional k-way approach and suggest how binary partitioning of attribute values may overcome them. We introduce a procedure for dichotomising sets of attribute values and show how this may be used to build binary trees using attributes with nominal values. The results of a comparative study of the performance of the proposed binary partition procedure and C4.5 demonstrate that the former sometimes performs markedly better and never performs significantly worse. Keywords: Binary Decision Trees, Zeta, Fragmentation 1 Introduction During the last two decades decision tree induction has become firmly established as the most widely used machine learning technique ([2], [...
