Saturated Neighborhood Graph Clustering Optimization Algorithm Based on Edge Information

Abstract

Clustering is a crucial task in the field of data mining. However, existing clustering algorithms often suffer from parameter dependence, which poses a challenge for users who need to possess a certain level of knowledge to effectively utilize these algorithms. This requirement raises the threshold for applying such algorithms. Furthermore, different clustering algorithms have varying parameter requirements, and the appropriate parameters may differ across different datasets. The introduction of non-parameter algorithms offers a promising solution to the criticized parameter tuning problem. In this study, some scholars have proposed the Saturated Neighborhood Graph (SNGC) clustering algorithm, which demonstrates excellent clustering effectiveness and possesses the characteristics of being non-parameter and robust. This study aims to optimize the clustering process of the SNGC algorithm and proposes a new clustering algorithm called SNGCE. The innovations of this research primarily encompass the following two points: 1) It addresses the issue of unconstrained during the construction of the saturated neighborhood graph; 2) It presents a novel approach to boundary detection within the clustering field. Experimental data indicates that the incorporation of boundary point information in this study can enhance the results of the clustering algorithm to a certain extent.