Abstract
Clustering is an old research topic in data mining and machine learning. There are so many clustering methods, and among them Wu-Huberman is few linear algorithm. In this paper, we propose the CBWH algorithm that does not only preserve the merit of Wu-Huberman, but also increase the robustness and extend the application of the Wu-Huberman. At first, we extend the Wu-Huberman to the general clustering problem by constructing graph. Then, we present a new idea to determine the threshold of the clustering. Additionally, we provide experiments for analyzing the effectiveness of the algorithm, comparing with other related algorithms and discussing the sensitivity, from which we find that CBWH is more robust.
Original language | English (US) |
---|---|
Pages (from-to) | 1531-1536 |
Number of pages | 6 |
Journal | ICIC Express Letters, Part B: Applications |
Volume | 3 |
Issue number | 6 |
State | Published - 2012 |
Externally published | Yes |
Keywords
- Clustering
- Data mining
- KNN (k nearest neighborhood)
- Wu-Huberman
ASJC Scopus subject areas
- General Computer Science