Hits:

Indexed by:会议论文

Date of Publication:2007-05-22

Included Journals:EI、CPCI-S

Volume:4426

Page Number:367-+

Key Words:data mining; clustering analysis; mechanics; minimum potential; energy principle

Abstract:Existing clustering algorithms use distance, density or concept as clustering criterion. These criterions can not exactly reflect relationships among multiple objects, so that the clustering qualities are not satisfying. In this paper, a mechanics based clustering algorithm is proposed. The algorithm regards data objects as particles with masses and uses gravitation to depict relationships among data objects. Clustering is executed according to displacements of data objects caused by gravitation, and the result is optimized subjecting to Minimum Potential Energy Principle. The superiority of the algorithm is that the relationships among multiple objects are exactly reflected by gravitation, and the multiple relationships can be converted to the single ones due to force composition, so that the computation can be executed efficiently. Experiments indicate that qualities of the clustering results deduced by this algorithm are better than those of classic algorithms such as CURE and K-Means.