提出一种基于数据挖掘的异常交易检测方法,可以在业务层面和操作层面对交易中的异常进行检测.当一个用户提交一笔新的消费交易时,采用贝叶斯信念网络算法判断当前交易属于正常交易的后验概率,作为在业务层面的可信因子；然后提取该用户在当前交易之前的若干个操作,与当前交易一起构成一个固定长度的操作序列,并通过BLAST-SSAHA算法将其与该用户正常操作序列和已知异常操作序列进行比对,得出在操作层面的可信因子.综合考虑业务层面的可信因子和操作层面的可信因子,最终决定当前交易是否为异常交易.%This paper presents a data mining-based transaction anomaly detection method, which can detect the fraud at both the business and operational level. When a user submits a new purchase request, the Bayesian belief network is used to determine the posterior probability that is the index of normal transaction, and use this as its credibility factor at business level. Then we extract user' s previous recen operations to form a fixed-length sequence of operations along with the current transaction. The BLAST-SSAHA aligns this sequence wit! user's normal operation sequences and known fraud operation sequences to arrive at the credibility factor at operational level. Considerin; comprehensively the credibility factors on these two levels, the final decision can be made on whether the current transaction is abnormal.
展开▼
