Aiming at the challenge of defining the subspace dissimilarity in traditional K-means soft subspace clustering techniques, this paper proposed a novel probability distance-based subspace difference representation model as the basis for an adaptive projection clustering algorithm. Firstly, based on the subspace clustering theory, the proposed method formulated a formula to describe the dissimilarities between associated soft subspaces. Secondly, by combining this formula with soft subspace clustering, it defined a clustering objective optimization function and provided a detailed clustering algorithm process according to a local search strategy. A series of experiments on both synthetic and real-world datasets demonstrate that the introduction of subspace comparison can lead to learning a more optimal soft subspace for clusters. Compared to existing mainstream subspace clustering algorithms, the proposed algorithm significantly improves the clustering accuracy, making it suitable for high-dimensional data clustering analysis.

[2023-06-01] Accepted Paper /doc/2023.03.0115/v1

[2023-11-05] Printed Article /doc/2023.03.0115/v2