一种保持最小广义汉明距的属性约简算法  

A Minimum Generalized Hamming Distance Algorithm for Attribute Reduction

作　　者：秦海鸥[1] 汤克明[1] 

机构地区：[1]盐城师范学院信息科学与技术学院,盐城224002

出　　处：《武汉理工大学学报》2009年第18期135-137,150,共4页Journal of Wuhan University of Technology

摘　　要：分析了当前一般属性约减算法的特点,给出了广义汉明距离、广义汉明距离矩阵、决策表最小广义汉明距的定义,在此基础上提出了一种保持决策表最小广义汉明距的属性约简算法。该算法对影响决策表最小广义汉明距的关键属性不进行约简,并约去非关键属性,从而有效保持了决策规则的抗噪声能力。仿真实验结果表明:与一般属性约减算法相比,保持决策表最小广义汉明距的属性约简算法的误识率显著下降。The features of current general algorithms for attribute reduction are analyzed. Generalized Hamming distance, generalized Hamming distance matrix and minimum generalized Hamming distance （MGHD） of decision table are defined, then a preserving MGHD of decision table algorithm for attribute reduction is proposed. The algorithm reduces the non-key attributes and the key attributes which influences the MGHD of decision table is preserved, so the anti-noise ability of decision rules is promoted. The simulation results show that compared with general algorithms the algorithm decreases the false recognition rate significantly.

关 键 词：最小广义汉明距 决策表 属性约简 抗噪能力 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]