A Graph Form Data Stream Clustering Approach Based on Dimension Reduction


Makul O., EKİNCİ M.

25th Signal Processing and Communications Applications Conference (SIU), Antalya, Türkiye, 15 - 18 Mayıs 2017 identifier identifier

Özet

In this paper, a new approach is presented for data stream clustering which is one of the popular subject in recent years. In this proposed approach, two distinct data stream algorithms are used. Proposed approach is based on integrating localized Linear Discriminant Analysis (LLDA) which is adopted from Linear Discriminant Analysis (LDA) for data stream to CEDAS which is used graph structure for clustering arbitrarily shaped clusters. LLDA is utilized dimension reduction method to make a projection to subspace consisting nearest clusters to data stream. First, cluster size are controlled in this approach. If cluster size is greater than threshold, clustering is (lone in the subspace obtaining from LDDA with CEDAS. Otherwise clustering is done in the full space by using CEDAS. Proposed approach is tested on CoverType, DS1 and Mackey-Glass data stream sets which arc commonly used in literature and achieved results demonstrate proposed approach's success.