一种改进的K-Modes聚类算法  

An improved K-Modes clustering algorithm

作　　者：石隽锋[1] 白妙青[1] 

机构地区：[1]山西大学计算中心,山西太原030006

出　　处：《现代电子技术》2015年第4期39-41,45,共4页Modern Electronics Technique

基　　金：云计算在山西农业生产信息化中的应用研究(2014041038-2)

摘　　要：传统的K-Modes算法采用0-1简单匹配方法计算对象与类中心(Modes)之间的距离,并将每个对象分配到离它最近的类中去。采用基于频率方法重新计算各类的类中心(Modes)、定义目标函数,然而,对象的归类方法和目标函数的定义没有充分考虑分类数据的特点。对此,提出一种改进的K-Modes算法,采用期望熵最小的衡量方法进行归类,并且采用期望熵作为新的目标函数。通过实验将该算法与传统的K-Modes算法进行比较,表明该算法是更有效的。The simple matching dissimilarity measure is used in the traditional K-Modes algorithm to calculate the distancebetween an object and cluster center,allocate each object to the nearest cluster,recalculate the cluster center of each cluster bymeans of the method based on frequency and define the objective function. However,the classification method for each datapoint and the definition of objective function don＇t consider the characteristics of categorical data fully. To solve the problem,an improved K-Modes algorithm is proposed,which adopts the minimum judging method of the expected entropy for classifica-tion,and takes expected entropy as a new objective function. The algorithm was compared with the traditional K-Modes algo-rithm in experiment. The result proved the effectiveness of the algorithm.

关 键 词：分类型数据 聚类算法 期望熵 目标函数 聚类精度 

分 类 号：TN911-34[电子电信—通信与信息系统] TP181[电子电信—信息与通信工程]