Feature Selection vs Theory Reformulation:A Study of Genetic Refinement of Knowledge-based Neural Networks
| 详细信息 | 推荐本文 | |
- 作者：Burns ; Brendan Davis ; Danyluk ; Andrea Pohoreckyj
- 关键词：theory refinement ; theory reformulation ; feature selection ; knowledge ; based neural networks ; genetic algorithms
- 刊名：Machine Learning
- 出版时间：2000
- 出版年：2000
- 期刊代码：22_08856125
- 类别：edu
- 卷：38
- 期：1/2
- 页码：89-107
- 数据来源：sp
摘要
Expert classification systems have proven themselves effectivedecision makers for many types of problems. However, the accuracy of suchsystems is often highly dependent upon the accuracy of a human expert'sdomain theory. When human experts learn or create a set of rules, they aresubject to a number of hindrances. Most significantly experts are, to agreater or lesser extent, restricted by the tradition of scholarship which haspreceded them and by an inability to examine large amounts of data ina rigorous fashion without the effects of boredom or frustration. Asa result, human theories are often erroneous or incomplete. To escapethis dependency, machine learning systems have been developedto automatically refine and correct an expert's domain theory.When theory revision systems are applied to expert theories, they oftenconcentrate on the reformulation of the knowledge provided rather than onthe reformulation or selection of input features. The general assumptionseems to be that the expert has already selected the set of features thatwill be most useful for the given task. That set may, however, be suboptimal.This paper studies theory refinement and the relative benefits of applyingfeature selection versus more extensive theory reformulation.