基于类内类间离散度的分类器设计方法.pdfVIP

: TP39 1 : A : 1009- 2552( 2010) 05- 0035- 03 1 2 郭亚琴 , 王正群 ( 1. , 226002; 2. , 225009) : 提出了一种基于类内类间离散度的最小距离分类器设计方法该方法解决了分类过程 中样本 分散和样本不可分问题基本思想是: 利用训练样本定义类内类间离散度矩阵, 根据 对离散度矩阵的分析, 建立目标函数, 求解目标函数, 得到一组最优解在分类时, 应用最优 解进行加权定义, 从而获得更好的识别结果采用 UCI标准数据集实验, 实验结果表明该算法 具有一定的优越性 : ; ; ; Classifier design based on between-class and w ithin-class scatter 1 2 GUO Y a-q in , WANG Zh ng-qun ( 1. Z ilang V ocational Technica l College, Nantong 226002, China; 2. School of Information Engineering, Yang hou University, Y ang hou 225009, China) Abstract: M inmi um d istanc classifi r bas d on b tw n-class andw ithin-class scatt r is propos d. Th n w m thod can r solv samp l disp rsion and th unclassifiabl r gion prob lm s. Th basic id a is to r solv an optmi al r solution. F irst, b tw n-class and w ithin-class scatt r matrix is d fin d by training samp l . Th n, obj ctiv function is s t up. Th optmi al w ight is obtain d through solving obj ctiv function. By adding w ight d fining for d istanc in th classification phas , th classifi r mi prov d its classification accuracy. Exp rmi nt is t st d on UCI standard databas . Th xp rmi ntal r sults show that th propos d m in mi um distanc classifi r bas d on b tw n-class andw ithin-class scatt r is ff ctiv . K ey words: classification; m in mi um distanc classifi r; b tw n-class and w ith in-class scatt r; optmi al r solution 0 引言 , , , , , , , , , , , K- , , , , , : 2009- 10- 20