Рақамли технологияларнинг назарий ва амалий масалалари 8-том 4-сан (2025) · 129-137-беттер
Analysis of Graph-based Cluster Core Separation using a new Core Separability Index
Адилова, Ф.Т., Давронов, Р.Р.
Аннотация
This article presents a new internal validity index of clustering called the Core Separation Index (CSI), designed to improve the quality and automate graph-based density clustering algorithms. The proposed index assesses the quality of cluster nucleus separation by measuring the ratio between the internal cohesion of each nucleus and its coupling with other nuclei. We present the theoretical basis of the CSI index, the theorem about its properties, and a strict proof. The index is integrated into the algorithm based on graph density variation analysis (GDVA), which allows for the automatic determination of optimal clustering parameters. A comprehensive experimental comparison of the CSI index with existing indices, such as DVI (density-based cluster validity indices), Silhouette, and Dunn, was carried out on five sets of synthetic data of different structures. The results, assessed using the external metrics Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI), show that CSI provides the best clustering quality (average NMI 0.7460), significantly exceeding other indices in calculation speed (average 0.17s). A high correlation of the proposed index with external quality indicators is also shown, which confirms its reliability for assessing and selecting the optimal structure of clusters.
кластерный анализкластеризация на основе графовплотностная кластеризацияиндекс валидности кластеровGDVACSIавтоматическое определение параметровARINMIcluster analysis
Метадайындар булагы: журналдын OAI-PMH архиви · Sindex толук текстти сактабайт, булакка шилтеме берет.