Scaling-up model-based clustering algorithm by working on clustering features

Abstract

In this paper, we propose EMACF (Expectation- Maximization Algorithm for Clustering Features) to generate clusters from data summaries rather than data items directly. Incorporating with an adaptive grid-based data summarization procedure, we establish a scalable clustering algorithm: gEMACF. The experimental results show that gEMACF can generate more accurate results than other scalable clustering algorithms. The experimental results also indicate that gEMACF can run two order of magnitude faster than the traditional expectation-maximization algorithm with little loss of accuracy. © Springer-Verlag Berlin Heidelberg 2002.