Incremental Induction of Topologically Minimal Trees
Abstract
This paper proposes the notion of ‘topological relevance’ as a means to formalize the complexity of a decision tree representation of a set of instances. It then presents an incremental algorithm, called IDL, for the induction of decision trees which are optimal according to this notion. IDL relies on ideas from ID4 and ID5 but searches using a statistical criterion for expanding nodes and a tree topological one, called topological relevance gain, for transforming decision trees. The results of its analysis and empirical validation show that, with provisions, IDL rapidly finds the same or a better tree than top-down induction algorithms and their incremental versions ID4, ID5 and ID5R.
Cite
Text
Van de Velde. "Incremental Induction of Topologically Minimal Trees." International Conference on Machine Learning, 1990. doi:10.1016/B978-1-55860-141-3.50012-2Markdown
[Van de Velde. "Incremental Induction of Topologically Minimal Trees." International Conference on Machine Learning, 1990.](https://mlanthology.org/icml/1990/develde1990icml-incremental/) doi:10.1016/B978-1-55860-141-3.50012-2BibTeX
@inproceedings{develde1990icml-incremental,
title = {{Incremental Induction of Topologically Minimal Trees}},
author = {Van de Velde, Walter},
booktitle = {International Conference on Machine Learning},
year = {1990},
pages = {66-74},
doi = {10.1016/B978-1-55860-141-3.50012-2},
url = {https://mlanthology.org/icml/1990/develde1990icml-incremental/}
}