A new approach to semi-supervised fuzzy graph-cut clustering

Abstract

Traditional feature-based clustering algorithms often fail to capture the geometric structure of data effectively. To address this limitation, this paper proposes a semi-supervised fuzzy graph cut clustering algorithm (SS-FGCC), which integrates anchor-point labels into fuzzy clustering and exploits graph structures to improve clustering quality while reducing computational complexity to linear time. Experiments on six synthetic datasets and five UCI datasets demonstrate that SS-FGCC consistently outperforms RCut, NCut, FCM, FFCAG, FGCC, and CEHM, achieving up to 82.76% accuracy. Moreover, the proposed method remains stable on large-scale datasets where conventional graph cut methods often suffer from memory limitations, making SS-FGCC an efficient and scalable clustering solution.