关键词: 高维数据, 聚类, 子空间, 信息熵, 层次聚类

Abstract: Subspace clustering is a kind of effective strategy to high-dimensional data clustering, the principle of subspace clustering is as well as possible keeping original data information, meanwhile as small as possible using subspace to data clustering. Based on the studying of the existing soft subspace clustering, it proposes a new algorithm for subspace searching. The algorithm combines with the size of cluster and information entropy, defines a new subspace dimensional weight distribution mode, and then uses the feature vector of cluster subspace to measure the similarity of two clusters. It uses the idea of agglomerative hierarchical clustering in hierarchical clustering to data clustering, which overcoming the shortcomings of using information entropy or traditional similarity separately. Through the test in the Zoo, Votes, Soybean three typical categorical data set to find out that compared with other algorithms, the proposed algorithm not only can improve the accuracy of clustering, but also has the very high stability.

Key words: high-dimensional data, clustering, subspace, information entropy, hierarchical clustering