Indexed by:期刊论文

Date of Publication:2011-07-01

Journal:Journal of Convergence Information Technology

Included Journals:EI、Scopus

Volume:6

Issue:7

Page Number:194-202

ISSN No.:19759320

Abstract:Support Vector Machine (SVM) is a prominent approach with adaptabilities in machine learning. It shows advantages in solving problems of pattern recognition. However, it is a bottleneck for SVM to cost a great deal of time when encountering a huge number of data. In this study, we present a policy of cluster analyzing applied to incremental SVM learning to improve SVM. Our model focuses on weight of time and distribution of data-features in cluster analysis and combines them with SVM to ensure the entire policy becomes an adaptive one for incremental data whose distribution of features shifts with time. Experimental results indicate that, under this condition, our policy could get better accuracy than those fast SVM algorithms which ignore the changes, and save time simultaneously. The proposed policy is thus practical and efficient.