改进的基于闵氏距离的软子空间聚类算法  

A new algorithm for Minkowski metric based soft subspace clustering

作　　者：支晓斌[1] 高垚琦 

机构地区：[1]西安邮电大学理学院,陕西西安710121 [2]西安邮电大学通信与信息工程学院,陕西西安710121

出　　处：《西安邮电大学学报》2015年第6期56-60,共5页Journal of Xi’an University of Posts and Telecommunications

基　　金：陕西省自然科学基金资助项目(2014JM8307);陕西省教育厅科学研究计划资助项目(14JK1661)

摘　　要：用L1范数和L2范数的加权组合取代基于闵氏距离的软子空间(Minkowski metric based soft subspace,MSC)聚类算法目标函数中所用的Lp范数,导出一个新的MSC的聚类中心计算公式,从而得出一种改进的MSC聚类算法。改进算法使MSC的计算复杂度由O(n2 mc)降为O(nmc)(这里n是数据个数,m为数据维数,c是聚类数)。在Iris,breastcancer,Vehicle,User和Wine 5个真实数据上的对比性实验结果显示,改进MSC算法的聚类精度与原MSC的聚类精度相当,但改进算法的运行时间是原MSC运行时间的1/7到1/2。An improved Minkowski metric based soft subspace （MSC） clustering algorithm is presented. A weighted combination of the L1 norm and L2 norm is used in the objective function of MSC clustering algorithm to replace the original Lp norm, and a new clustering center calculation formula of MSC is therefore derived. The improved MSC clustering algorithm can reduce the computational complexity of MSC from O（n2mc） to O（nmc） （where n is the number of data, m is the dimension of data, c is the number of clusters）. Experiments results on 5 real data sets including Iris, Breastcancer, Vehicle, User and Wine algorithm has competitive clustering precision, but the running between seventh and one half of the time by original MSC. show that the improved MSC time of the improved algorithm is

关 键 词：聚类算法 软子空间 闵科夫斯基距离 加权组合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]