基于相似度的离群模式发现模型

提出了基于相似度的离群模式发现模型，该模型主要利用知识属性集分析离群点，既能够处理离群点的数值属性，又能够处理其类别属性；通过组间相似度从中发现离群模式，不仅回避离群点数量少的缺陷，还利用了离群点的隐含语义．给出了在银行结售汇交易数据上进行的实验分析结果，模型发现了某地区的3个可疑模式，该结果为金融犯罪分析提供有利线索；利用不同子空间角色划分，可以发现个人、地区等不同对象间的异常资金流动；模式发现算法具有线性时间复杂度，在实际应用中具有较好的性能．结果表明模型能检测出可疑资金流动序列，为反洗钱工作提供有意义的线索．

The model of outliered pattern mining based on similarities

This paper gave a model for mining outliered patterns. It used knowledge sets to deal with both categorical and behavioral property of outliers and mined outliered patterns based on group similarities to avoid the lack of outliers and make use of their intentional knowledge. The expermental result of its application in the foreign concurrency transaction data of bank was presented. The model detect three suspicious pattens of an area which is a good hint for financial crime detection; using different subspace role assignment, the model can detect outliered fund flows between persons and areas; the algorithm is of linear time complexity and has a good performance in application, it can detect suepiaous fund flow and provide significant thread for auti money laundering.