局部PCA与k近邻相结合的谱聚类算法

为解决传统谱聚类方法构造相似矩阵时使用距离度量无法充分反映复杂样本空间的全局一致性,且存在聚类个数需要人为指定等问题,提出一种基于近邻与局部PCA结合的谱聚类算法。使用kNN获取邻域子集,使用局部主成分分析保持样本的局部结构,同时考虑样本的全局和局部信息,为相似矩阵提供综合信息,对得到的相似矩阵进行图分割,直接得到聚类的结果。在真实数据集上的实验结果表明,该算法能够自动得到类的个数且聚类准确率有所提高,其在非真实类时有更好的聚类效果。

Spectral clustering algorithm combining local PCA with k nearest neighbor

To solve the problems that the traditional spectral clustering method cannot fully reflect the global consistency of the complex sample space when constructing similarity matrix and the number of clusters needs to be artificially specified,a method based on the combination of neighbor and local PCA was proposed.The kNN was used to obtain the neighborhood subset and the local principal component analysis was then used to maintain the local structure of the sample.In this way,both the global and local information of the sample were considered,which provided comprehensive information for the similarity matrix.The obtained similarity matrix was transformed into a graph to directly obtain the clustering result.Experimental results on real data sets show that the proposed algorithm can automatically obtain the number of classes and the clustering accuracy can be improved.It also shows better clustering effects in other classes.