A Hill-Climbing Approach to Construct Near-Optimal Decision Trees

Abstract

We consider the problem of identifying the state of an $n$ component coherent system, where each component can be working or failed. It is costly to determine the states of the components. The goal is to find a decision tree which specifies the order of the components to be tested with minimum expected cost. The problem is known to be NP-hard. We present an extremely promising heuristic method for creating effective decision trees, and computational results show that the method obtains optimal solutions for $95 %$ of the cases tested.

Cite

Text

Sun et al. "A Hill-Climbing Approach to Construct Near-Optimal Decision Trees." Pre-proceedings of the Fifth International Workshop on Artificial Intelligence and Statistics, 1995.

Markdown

[Sun et al. "A Hill-Climbing Approach to Construct Near-Optimal Decision Trees." Pre-proceedings of the Fifth International Workshop on Artificial Intelligence and Statistics, 1995.](https://mlanthology.org/aistats/1995/sun1995aistats-hillclimbing/)

BibTeX

@inproceedings{sun1995aistats-hillclimbing,
  title     = {{A Hill-Climbing Approach to Construct Near-Optimal Decision Trees}},
  author    = {Sun, Xiaorong and Chiu, Steve Y. and Cox, Louis Anthony},
  booktitle = {Pre-proceedings of the Fifth International Workshop on Artificial Intelligence and Statistics},
  year      = {1995},
  pages     = {513-519},
  volume    = {R0},
  url       = {https://mlanthology.org/aistats/1995/sun1995aistats-hillclimbing/}
}