分类数据集的一致化特征选择约简  

Consistent feature selection reduction about classification data set

作　　者：吴新玲[1,2] 

机构地区：[1]广东技术师范学院信息工程系 [2]武汉大学软件工程国家重点实验室,武汉430072

出　　处：《计算机工程与应用》2007年第18期174-176,共3页Computer Engineering and Applications

基　　金：武汉大学软件工程国家重点实验室开放基金资助(No.SKLSE05-09)。

摘　　要：样本数据集的不一致性和冗余特征会降低分类的质量和效率。提出了一种一致化特征选择约简方法,该方法基于贝叶斯公式,采用阈值,将非一致数据归为最可能的一类,使数据集一致化。并在一致数据集上,运用类别区分矩阵选择可准确区分各类数据的最小特征变量集。给出的启发式搜索策略和应用实例表明:一致化特征选择约简方法能有效消除分类数据集的不一致性,选择最优的特征变量、降低数据的维数、减少数据集中的冗余信息。The disaccords and the redundancy features of a sample dataset will drop the classification quality and efficiency. In this paper,the method called consistent feature selection reduction is proposed about the c]assification data set.This method group together the inconsistent datum of the best possible category and make the data set uniform based on the Bayesian formula and a threshold value.Then a category distinguish matrix is built upon the consistent data set and the least feature variable subset that can distinguish the classification accurately is obtained through the category distinguish matrix.A heuristic search strategy and a practical example are given.The result shows the consistent feature selection reduction method can eliminate the disaccords of the sample dataset,select the optimal feature variables,drop the dimension of the data and reduce the redundancy information effectively.

关 键 词：数据挖掘 分类 特征选择 数据约简 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]