基于相似性保和特征变换的高维数据聚类改进算法.PDFVIP

40 3 Vol.40, No. 3 20 11 6 Acta Geodaetica et Ca rtographica Sinica Jun. , 20 11 :(2011) 1 1, 2 1 1 王家耀 , 谢明霞 , 郭建忠 , 陈 科 1. , 4 50052; 2. 757 19 , 430074 Improved High Dimen ional Data Clu tering Algorithm Ba ed on Similarity Pre erving and Feature Tra n formation 1 1,2 1 1 WANG Jiayao , X IE Mingxia , GUO Jianzhong , CHEN Ke 1. In titute of Surveying a nd Mapping, Information Engineering Univer ity, Zhengzhou 450052, China ; 2. 757 19 Troup, Wuha n 430074, China Ab tract : Improved high dimen ional data clu tering a lgorithm ba ed on imilarity pre erving a nd fea ture tra n for ma tion i propo ed. Fir tly, gain the imila rity mat rix of high dimen iona l data with the de igned imila rity mea ure function, and tran la te it into di ta nce matrix of the horte t path through the neare t neighbor earching met hod and the algorithm Floyd. Then, tran la te high dimen iona l feature tra n format ion into the optimiza tion a nd re olve thi optimiza tion problemwith genetic algorithm. The reduced da ta i u ed for clu tering analy i via kmean and the va lue pa ir between the coordinate of high dimen iona l data and their reduced 2D coordinate are u ed for RBF neural network training. Determine the belongingne of new object ba ed on the di ta nce from the new object to each current clu tering center through the trained neural network. Finally, the experimental re ult provethe validity of the improved imilarity mea ure a nd the operability of the dimen ionality reduction method ba ed on fea ture tran formation. Key word : feature tran formation; high dimen iona l data clu tering; imila rity mea ure; dimen ionality reduct ion :提出一种基于相似性保持和特征变换的高维数 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