Binary Information Press
High dimensional data clustering is an important issue for data mining. Firstly, the records in the dataset are mapped to the vertices of hyper-graph, the hyper-edges of hypergraph are composed of the vertices which have the same value on one certain attributes. Then, a multilevel hypergraph partitioning algorithm is used to find k parts of the hypergraph. Finally, the clusters with good quality are chose out by computing the value of clusters validity. The experimental results show that the new algorithm can efficiently construct high quality clusters.