摘要
为改善近邻传播聚类算法对高维数据的聚类效果,引入马氏距离替换原算法中的欧氏距离,并借助正则化总散度矩阵的奇异值分解实现数据变换预处理,进而在在降维后的变换子空间中对数据集进行聚类。针对Iris、User、Soybean和Vehicle四个数据集,选取适当正则化参数,经仿真实验可见,改进算法的聚类精度在整体上有所提高。
In order to improve the affinity propagation clustering algorithm for high dimensional data clustering effect,the Mahalanobis distance is introduced to replace the original Euclidean distance,a data transformation preprocessing is carried out by means of singular value decomposition of the regularized total scatter matrix,then,the data sets are clustered in the transformed subspace after dimension reduction.For the four data sets of Iris,User,Soybean and Vehicle,the appropriate regularization parameters are selected,and the simulation results show that the clustering accuracy of the improved algorithm is improved on the whole.
出处
《西安邮电大学学报》
2017年第6期46-49,共4页
Journal of Xi’an University of Posts and Telecommunications
基金
国家自然科学基金资助项目(61671377)
陕西省自然科学基金资助项目(2014JM8307)
关键词
近邻传播聚类
高维数据
马氐距离
正则化
affinity propagation clustering
high dimensional data
Mahalanobis distance
regularization
