Granular-ball-based Fast Spectral Embedding Clustering Algorithm for Large-Scale Data

When dealing with large-scale datasets, manifold learning, which is crucial for analyzing high-dimensional data, faces challenges, including low clustering accuracy and high computational complexity. In this paper, we introduce granular-ball into unsupervised manifold learning. Based on the anchor graph generated by granular-ball and spatial information, we propose a novel granular-ball-based fast spectral embedding clustering algorithm, named GB-FSEC. The GB-FSEC algorithm first employs a strategy combining granular-ball and random sampling to generate representative anchor points, constructing a new adjacency matrix to reduce data dimensionality significantly, thereby lowering computational complexity. Moreover, to avoid the complexity of adjusting the kernel parameter, GB-FSEC adopts a non-parametric strategy. Experimental results demonstrate that, compared to other methods, this approach can handle large-scale datasets and exhibits excellent performance in terms of clustering accuracy.