摘要： 计算信息特征(属性)的权重问题在信息分类及模式匹配中是一个研究热点。该文提出一种基于改进ReliefF算法的主成分特征提取方法，利用此算法删除原始特征中与分类不相关的特征，并对数据进行归一化处理和主成分提取。实验将34个特征变量降维成10个主成分，大大减轻后续的分类器工作量，提高分类器的分类精度。

关键词: ReliefF算法, 特征提取, 主成分分析

Abstract: To compute the weight of information feature is a very important research work in information classify and pattern matching, a method of principal feature selection based on reformative algorithm ReliefF is presented in this paper. The main work as fellow: reformative ReliefF algorithm is used to take out some irrelevant features from originality features, normalization is worked, and the principal feature is selected. Using the method to select the principal feature will reduce the dimensionality significantly. Experiment provides that, using the method, only 10 principal features is remained from 34 originality features, and the computing cost of the classify program can be decreased, at the same time, the precision of the classify program can be increased.

Key words: ReliefF algorithm, feature selection, principal component analysis

中图分类号: 

• TP391