基于相似性保持和特征变换的高维数据聚类改进算法  

Improved High Dimensional Data Clustering Algorithm Based on Similarity Preserving and Feature Transformation

作　　者：王家耀[1] 谢明霞[1,2] 郭建忠[1] 陈科[1] 

机构地区：[1]信息工程大学测绘学院 [2]75719部队

出　　处：《测绘学报》2011年第3期269-275,共7页Acta Geodaetica et Cartographica Sinica

基　　金：国家863计划(2009AA12Z228);国家科技支撑计划课题(2007BAH16B03)

摘　　要：提出一种基于相似性保持和特征变换的高维数据聚类改进算法。首先,通过相似性度量函数计算得到高维空间对象相似度矩阵,并利用近邻法、Floyd最短路径算法将相似度矩阵转换为最短路径距离矩阵;然后,将高维特征变换转化为遗传优化问题,利用特征变换降维后的二维数据进行k-均值聚类,并根据(高维坐标,降维后二维坐标)值进行RBF神经网络训练,当新对象输入时,利用训练好的神经网络对其进行二维映射,通过判断该对象与各聚类簇中心距离的远近获得其归属;最后,通过试验验证了改进相似性度量函数能够有效表达高维数据对象间的相似性,且基于特征变换的降维方法具有可操作性。Improved high dimensional data clustering algorithm based on similarity preserving and feature transformation is proposed.Firstly,gain the similarity matrix of high dimensional data with the designed similarity measure function,and translate it into distance matrix of the shortest path through the nearest neighbor searching method and the algorithm Floyd.Then,translate high dimensional feature transformation into the optimization and resolve this optimization problem with genetic algorithm.The reduced data is used for clustering analysis via k-means and the value pairs between the coordinates of high dimensional data and their reduced 2D coordinates are used for RBF neural network training.Determine the belongingness of new object based on the distance from the new object to each current clustering center through the trained neural network.Finally,the experimental results prove the validity of the improved similarity measure and the operability of the dimensionality reduction method based on feature transformation.

关 键 词：特征变换 高维数据聚类 相似度 降维 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]