归纳式学习中连续型数据的区间划分问题

主要研究的是归纳式学习中连续型样本数据的分组问题 .在归纳式学习方法中 ,对样本数据的不同分组会直接影响到所生成的反映变量间相互关系的规则 ,从而影响到对新样本的识别效果 .特别是对样本数据为连续型的数据的分组问题 ,目前尚未得到很好的解决 .本文在类相关离散化方法的基础上 ,提出了用极大熵法进行初始区间划分 ,用多因素优选法调整区间的边界 ,二阶概率统计检验与实际意义相结合进行区间合并的一整套划分区间的方法 ,并用本文建立的新方法体系对中国宏观经济的某些指标间的相互关系进行了分析 ,取得了良好的效果.

Discretization for Inductive Learning from Continuous Sample Data

This paper is mainly about discretization for inductive learning from continuous sample data. In inductive learning approach, how to discretize the sample data can directly affect the creation of the production rules and the recognizing result. But the problem of discretization of continuous data hasn't been well solved. In this paper, a new discretization method, based on the class\|dependent discretization, is proposed. This approach consists of three main processes: interval initialization using the maximum entropy criterion, interval improvement using multi\|dimensional optimization, and interval combination using the second\|order probability statistical estimation and the criteria from real world. In addition, this paper uses the new method to analyze some problems of China macro\|economy. It is proved that the method in this paper can get a promising result.