Fast adaptive clustering for very large datasets
Kun-Che Lu, Don‐Lin Yang, Jungpin Wu · 2005
Abstract:- In this paper, we propose an efficient and effective clustering method that requires to scan a dataset only once. The original dataset is transformed first and merged into a hyper image of controllable size. Unlike traditional methods, the dissimilarity measurement between objects is calculated once for all objects by using various image processing methodologies, such as morphological operations. Image connect component extraction is thereby used to extract clusters from the hyper image. The proposed method is easy to use for clustering data in way of fuzzy and hierarchical fashion readily under a single dataset scan. It is also efficient for incremental and dynamic clustering without additional scan of the original dataset. Experimental results show that the proposed method is robust and stable under various parameter settings such that it is more effective and useful than traditional clustering methods, especially for very large datasets.