一种新的基于局部保持投影的高维数据聚类成员构造方法

摘要

研究在高维数据中如何产生聚类成员,并提出一种新的构造聚类成员的方法.为解决高维数据的维度对构造成员带来的影响,新的构造方法在构造聚类成员之前利用局部保持投影先对高维数据进行维度约减,然后在约减后的子空间中用随机投影结合K均值方法构造聚类成员.最后讨论了局部保持投影子空间维度的选取.实验表明,新方法得到的结果要明显优于已有的主分量分析结合下采样方法和简单的随机投影方法.%This paper studied how to construct cluster ensembles for high dimensional data and proposed a new ensem ble constructor. To ameliorate the effect caused by high dimensionality;the proposed method used Locality Preserving Projections(LPP) to reduce the dimensionality before constructing ensembles. Then constructed ensembles based on random projection combined with K means in LPP subspace. Finally;we discussed how to choose the dimensionality of LPP subspace. The experiments show that ensembles generated by new algorithms perform better than those by Princi pal Component Analysis with subsampling(PCASS) and simple Random Projection(RP) that was proposed before.