ID5 and ID5R (Utgoff, 1989) are both incremental decision tree builders that overcome the deficiencies ID4. The essential difference is that when tree restructuring is required, because the attribute at a node does not have the lowest entropy score, any sub trees are not discarded, rather the attribute that is to be placed at the node is pulled up to the node and the tree structure below the node is retained.
In the case of ID5 the sub trees are not recursively updated while in ID5R they are. Not restructuring the sub trees is computationally more efficient. However, the resulting sub tree is not guaranteed to be the same as the one that would be produced by ID3 on the same training instances. ID5R does guarantee this to be the case.
Authored by Serengul Smith
E-mail to:
serengul1@mdx.ac.uk
School of Computing Science Middlesex University
Revised: September 1998