Document Type : Original/Review Paper
Authors
1 Departments of computer engineering, Tecnical and Vocational University (tvu), tehran, iran.
2 Department of Computer Engineering at the Technical and Vocational University (TVU), Tehran, Iran.
Abstract
With the continuous advancement of technology, enormous volumes of data are generated every moment. Extracting meaningful insights from this data to support decision-making across various fields has become a major area of research. Clustering serves as a fundamental technique for exploring and managing large datasets. However, traditional density- and neighborhood-based clustering algorithms, such as Fast Search and Density Peak Clustering and DBSCAN, often struggle to accurately identify clusters due to their reliance on limited neighborhood ranges and the inherent difficulty of simultaneously accounting for both global and local relationships during density estimation. To address these limitations, the proposed method introduces a three-phase strategy designed to establish clear distinctions between central and peripheral data points. This technique begins by selecting seed nodes along with their most closely related neighbors, determined through density coefficients, and then identifies strong internal linkages through a weighted network clustering process. By incorporating local neighborhood radius metrics to analyze density patterns and applying weighted clustering coefficients to assess similarity, the method enhances precision. Experimental evaluations on both real-world and synthetic datasets demonstrate that the proposed approach delivers more accurate clustering outcomes than many leading existing methods. On average, the proposed method achieved an ARI of 0.697 across 12 real-world datasets and demonstrated overall competitive performance compared to the benchmark methods.
Keywords
Main Subjects
)