A Heuristically Weighting K-Means algorithm for subspace clustering

Soft subspace clustering algorithms receive wide interests recently, because of their scalable and flexible ability at handling high dimensional sparse data. A disadvantage of those existing algorithms is their clustering results are affected by goodness of initial centroid selected by random initial method greatly. In this paper, we propose a heuristically weighting K-means algorithm and a corresponding initial method for clustering high-dimensional data. Experimental results have shown its effectiveness and stability.