Bayesian network-Based Projected Clustering

Clustering in high dimensional data is an important task. Projected clustering has emerged as a possible solution to the challenges associated with high dimensional clustering. A projected cluster is a subset of points together with a subset of attributes, such that some category of value of cluster points has great probability in these attributes. The relationship between projected cluster and Bayesian network is discussed in this paper. The dense cells can be identified by Bayesian network, and then adjacent dense cells can be merged into a projected cluster. It avoids the blind selection of dimension and the chicken-and-egg problem that must identify dimensions and data objects simultaneously. It can also search clusters on arbitrary subspace. The experiment results on categorical attributes show that Bayesian network-based projected clustering performs well in high dimensional data.