关键词: 数据挖掘, 离群点, 聚类, 超图分割

Abstract: Traditional mining algorithm does not contain relations which is defined in pattern. The clustering method based on traditional data pattern brings in different businesses together. Thus the result of the cluster is not good enough. In this paper, a new algorithm called PCOT (patternbased clustering outlier test) is presented. PCOT is suitable in highdimensional space, which uses a new business containing pattern. In the algorithm, a novel hypergraph model is proposed to represent the relations among the patterns. Hypergraph partitioning method is used in clusering. Experiment shows that this approach can find the outliers in highdimensionalsparse space effectively.

Key words: data mining, outlier, clustering, hypergraph partitioning