An Experimental Comparison of Three Kinds of Clustering Algorithms

Clustering is one of the most important and well studied fields of data mining. Many clustering methods have been proposed in the last few decades with different backgrounds. An open problem in clustering field is the lack of a benchmark for contrasting all these algorithms as well as other available ones. In this paper, we use the well-known log marginal likelihood (LML) score and classification accuracy as two criteria of comparison. Experiment results on k-means, expectation maximization (EM) and farthest-first in the t-test show that EM outperforms other two algorithms on most of the benchmark data sets with respect to both criteria. Considering the performance and comprehensibility of EM, it is an ideal model that could be used in many real world applications