基于属性权重最优化的k-means聚类算法  

A k-means Algorithm Based on Optimization of Attribute Weight

作　　者：熊平[1] 顾霄[1] 

机构地区：[1]中南财经政法大学信息与安全工程学院,湖北武汉430074

出　　处：《微电子学与计算机》2014年第4期40-43,共4页Microelectronics & Computer

基　　金：国家自然科学基金项目(71173238);教育部人文社科研究青年基金项目(12YJC630078)

摘　　要：聚类是最常用的数据挖掘算法之一.为了提高聚类结果的质量,应用拉格朗日乘数法提出了一种基于属性权重最优化的k-means聚类算法.该算法在计算样本与质心的距离时为各属性赋予相应的权重以表示属性的重要程度,并在每轮迭代中根据质心向量的变化自动计算最优的属性权重,使得所有样本与相应质心的距离和最小.实验结果验证了该方法相对于传统k-means算法的优势.Clustering is one of the most widely used data mining algorithm. In order to improve the quality of clustering results, we present a k-means algorithm based on optimization of attribute weight using Lagrange multiplier method. The algorithm uses the weight value for each attribute to determine the importance of that attribute while computing the distance between an instance and the centroid of a cluster. In each iteration of clustering, it computes the optimal weight of attributes according to the change of centroid vector which minimizes the sum of distance between each instance and the centroid. The experimental results demonstrate that the quality of clustering results can be improved significantly with the algorithm comparing with the traditional k-means algorithm.

关 键 词：聚类算法 属性权重 数据挖掘 目标函数 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]