We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Versatile Hyper-Elliptic Clustering Approach for Streaming Data Based on One-Pass-Thrown-Away Learning.
- Authors
Wattanakitrungroj, Niwan; Maneeroj, Saranya; Lursinsap, Chidchanok
- Abstract
Finding patterns or clusters in streaming data is very important in the present information mining. The most critical issue is the huge amount of data versus the limited size of storage space. In the previous works, the essential information of huge data was represented by subsets of data, grid summarization, or spherical function. Those forms of data representation are not compact enough to capture the topology of the arriving data points and may lead to the lack of information for generating the accurate cluster result. In this work, we proposed a new versatile hyper-elliptic clustering algorithm, called VHEC, to cluster the streaming data in one-pass-thrown-away fashion in order to preserve the original topology of data space. To cope with the problem of one-pass-thrown-away clustering, a new set of elliptic micro-cluster parameters, i.e. boundary, density, direction, intra-distance and inter-distance, was introduced. Furthermore, a feasible technique for merging two micro-clusters was developed. The proposed parameters and one-pass-throw-away clustering algorithm were tested against several benchmark data sets and structural clustering data sets. Our performance was compared with existing algorithms. Regardless of different sizes, shapes, and densities, VHEC outperformed the other previous data stream clustering algorithms on both synthetic and real data sets. Moreover, VHEC is more significantly robust to streaming speed and incoming data sequence than the other compared algorithms in terms of purity, Rand index, and adjusted Rand index measures.
- Subjects
MICROCLUSTERS; SPHERICAL functions; TOPOLOGY; ALGORITHMS; RAND index
- Publication
Journal of Classification, 2017, Vol 34, Issue 1, p108
- ISSN
0176-4268
- Publication type
Article
- DOI
10.1007/s00357-017-9222-1