基于簇间距离自适应的软子空间聚类算法

针对软子空间聚类过程中簇间距离(簇间的分离程度)对聚类的影响程度不确定的问题,提出了一种基于簇内紧密度和簇间距离自适应软子空间聚类算法。算法以经典的k均值聚类算法框架为基础,在最小化各个子空间簇类的簇内紧密度的同时最大化各个子空间簇类的簇间距离。并且通过推导得到新的子空间聚类中心和特征加权的计算方式,克服了软子空间聚类对输入参数敏感的缺点,实现了算法的自适应学习,并且取得了较好的聚类效果。

Soft subspace clustering algorithm based on self-adaption of intercluster distance

For the uncertain problem that intercluster distance（intercluster separation） influences on clustering in the soft subspace clustering process, a self-adaptive soft subspace clustering algorithm has been proposed based on the compactness of intracluster compactness and the intercluster distance. Minimize the intracluster compactness, and meanwhile maximize the intercluster distance based on the framework of classical k-means clustering algorithm. And a new way of computing clusters’ centers and features weighting is gotten by derivation. This way overcomes the sensitive defect of input parameters, realizes the self-adaptive learning, and obtains better clustering results.