基于时间序列建模和控制图的异常交易检测方法

可疑交易识别是打击洗钱犯罪所要面对的一项重要任务.为辅助反洗钱分析人员从海量金融交易信息中甄别客户异常交易,本文提出一种新的基于非线性马尔科夫随机过程、相空间重构和隐马尔科夫链的非线性随机方法,用于对金融交易时序进行建模拟合,然后应用鲁棒控制图对估计误差进行检验以发现异常.应用该算法对实际交易数据和仿真数据的分析验证了所提方法的有效性和可行性,可以被用于异常交易的监测.

An Approach for Unusual Transaction Detection Based on Time Series Modeling and Control Chart

Detecting suspicious transactions is a vital task for fighting against money laundering. To help anti-money laundering analyst screen customer's unusual transactions and behaviors in massive financial transaction information,we propose a nonlinear stochastic approach based on nonlinear Markov stochastic process,phase space reconstruction and hidden Markov chain for Modeling and fitting financial transaction time series.Then a robust control chart is applied to the estimation of errors from the fitting to detect anomalies.Applying the algorithm to real data examples and simulation,the experiment results suggest that the approach is effective and feasible and can be used for helping the detection of unusual transaction.