查看文章

In real life data sets some attributes may have lower importance or even may be completely noninformative. The subspace clustering algorithms have been proposed to handle this. The soft subspace algorithms are vulnerable to noise and outliers. The paper presents a novel algorithm that handles both various importance of attributes and outliers. The proposed Fuzzy Weighted C-Ordered Mean (FWCOM) clustering algorithm elaborates clusters in soft subspaces. In each cluster each attribute is assigned a weight from interval [0, 1]. Each attribute has its individual weight (importance) in each cluster. The proposed algorithm applies the ordering technique to effectively reduce the influence of outliers and noise. The paper is accompanied by numerical experiments.