摘要： 通过改进层次聚类和k-means聚类，建立两阶段聚类方法。采用两阶段聚类识别出异常点并得到高质量的聚类结果。结合证券公司客户真实交易数据和人工数据，使用Clementine进行建模从而实现聚类过程，识别出异常值并计算可疑记录的可疑程度，为金融情报部门提供了高质量的调查数据。

关键词: 层次聚类, k-means聚类, 数据挖掘, 可疑交易, 洗钱

Abstract: This paper improves the hierarchical and k-means clustering, builds the two-stage clustering method. It gets the outliers and high-quality clustering results by the two-stage clustering method. It uses Clementine to model the process of realization of clustering by clients’ real transaction records in securities companies and manual data, identifies outliers and calculates the suspicious degree of the records, and provides high-quality survey data for the financial intelligence departments.

Key words: hierarchical clustering, k-means clustering, data mining, suspicious transactions, money laundering

中图分类号: 

• TP391