G-TREACLE: A new grid-based and tree-alike pattern clustering technique for large databases

Abstract

As data mining having attracted a significant amount of research attention, many clustering methods have been proposed in past decades. However, most of those techniques have annoying obstacles in precise pattern recognition. This paper presents a new clustering algorithm termed G-TREACLE, which can fulfill numerous clustering requirements in data mining applications. As a hybrid approach that adopts grid-based concept, the proposed algorithm recognizes the solid framework of clusters and, then, identifies the arbitrary edge of clusters by utilization of a new density-based expansion process, which named "tree-alike pattern". Experimental results illustrate that the new algorithm precisely recognizes the whole cluster, and efficiently reduces the problem of high computational time. It also indicates that the proposed new clustering algorithm performs better than several existing well-known approaches such as the K-means, DBSCAN, CLIQUE and GDH algorithms, while produces much smaller errors than the K-means, DBSCAN, CLIQUE and GDH approaches in most the cases examined herein. © 2008 Springer-Verlag Berlin Heidelberg.