异常金融交易行为模式识别中的特征提取

Feature Extraction in Abnormal Pattern Recognition of Financial Transactions

特征提取的目的是获得能够被机器识别的数学特征。区别于传统的金融时间序列的特征提取与相似性度量方法,提出了一种基于径向基函数RBF神经网络一步预测误差序列特征提取与相似性度量方法。该方法将时间序列之间的相似性度量换化成特征矢量之间的相似性度量,并且建立了特征矢量与物理信息的关联,能够有效的检测出异常的金融交易行为模式。实验证明该方法相对于传统的直接距离、傅立叶变换、ARMA模型法具有明显优势。

Feature extractors are used to get mathematical features that can be machine -readable.In this paper a novel feature extraction and similarity measurement method based on Radial Basis Function neural network one-step deviation prediction is proposed,which is different from traditional time series researches.The method converts time series similarity to feature vectors similarity comparison,while feature vectors are associated with physical information.Experiments show that this method has obvious advantages compared to traditional time series researches like direct distance,Fourier transform,ARMA model method.It can effectively detect abnormal patterns of financial transactions.