Abstract: Outliers usually contain important information, it can help improving the users' understanding of the data.New definitions of cause attribute subspace of outliers, degree of cause attribute subspace and similarity of outlicrs were given, and then an algorithm for finding intentional knowledge of outliers based on attribute subspace was proposed, the approach can obtain the cause attributes set of every outlier. Then the outliers were classified by their similarity combined with the thinking of clustering, all the outliers of every class have the same cause attributes set under certain precision. The experiment results show that the algorithm is effective and practical,and more ease of use.

Key words: Outlicrs,Attribute subspace,0utlicrs similarity,Intentional knowledge