基于RBF神经网络的可疑交易监测模型

Suspicious transaction detection model based on Radial Basis Function Neural Network

针对国内外金融领域可疑交易的低检测率问题,通过对RBF(Radial Basis Function)神经网络技术的分析与研究,提出了一种基于APC-III聚类算法和RLS(Recursive Least Square)算法的面向反洗钱的RBF神经网络模型并加以实现。APC-III聚类算法用于确定RBF神经网络隐含层的中心向量,RLS算法用来调整隐含层与输出层之间的连接权值。RBF神经网络与支持向量机(SVM)和孤立点检测相比,有更高的检测率和较低的误检率,因此,提出的模型具有重要的理论和实用价值。

Aiming at the low detection rate of suspicious transaction at home and abroad in financial field,and with the analysis of Radial Basis Function（RBF） Neural Network,a RBF Neural Network model based on APC-Ⅲ clustering algorithm and Recursive Least Square（RLS） algorithm for anti-money laundering is proposed.APC-Ⅲ clustering algorithm is used for determining the parameters of RBF in hidden layer,and RLS algorithm is adopted to update weights of connections between hidden layer and output layer.The proposed method is compared against Support Vector Machine（SVM） and outlier detection methods,which show that the proposed method has the highest detection rate and the lowest false positive rate.Thus the model is proved to have both theoretical and practical value.