A Semi-Supervised Spectral Clustering Algorithm Based on Rough Sets

Abstract

Spectral clustering algorithm is an increasingly popular data clustering method, which derives from spectral graph theory. Spectral clustering builds the affinity matrix of the dataset, and solves eigenvalue decomposition of matrix to get the low dimensional embedding of data for later cluster. A semi-supervised spectral clustering algorithm makes use of the prior knowledge in the dataset, which improves the performance of clustering algorithms. In the paper, a semi-supervised spectral clustering algorithm based on rough sets is proposed, and extends rough set theory to the spectral clustering. The algorithm makes the clustering into a two-tier structure of upper and lower approximation, which can be used to settle the overlapping phenomenon existing in the dataset. Experiment proved that compared with existing algorithms, the modified algorithm obtains a better clustering performance.