基于深度卷积网络的可疑交易识别

Suspicious Transaction Recognition Based on Deep Convolutional Networks

可疑交易识别是反洗钱工作的重要内容,以算法模型为工具分析和识别可疑交易已成为新的趋势。深度卷积神经网络可有效自动提取数据中的分类特征,在众多分类任务中表现出较好的识别效果,已被广泛应用于各领域研究。本文首先基于深度学习理论,选取一维卷积神经网络,并设计了包含7层的模型框架应用于可疑交易识别分析。其次,将Elliptic数据集以7:3的比例划分为训练集和测试集,采用划分的数据对模型进行训练和测试,以GCN模型、Skip-GCN模型、EvolveGCN模型等深度神经网络模型为对照组,验证本文所提出模型的有效性。最后,通过将输入数据中各元素的排序方式随机打乱,探讨模型对数据输入的稳健性。研究结果表明,一维卷积神经网络对可疑交易识别具有较好的适用性,Elliptic数据集总体分类精度可达98%,F1值达到80%,具有较好的分类效果。对比GCN模型、Skip-GCN模型、EvolveGCN模型等在Elliptic数据集上的识别效果,本文所提模型总体上具有较好的识别精度,总体正确率和F1值均达到较高水平。

The identification of suspicious transactions is an important part of anti-money laundering work.The use of algorithms as tools to analyze and identify suspicious transactions has become a new trend.The use of deep convolutional neural networks could effectively extract classification features in data automatically.It has shown a good recognition effect in many classification tasks and has been widely used in various fields of research.The research steps of this article were as follows:Firstly,based on deep learning theory,this paper selected a one-dimensional convolutional neural network,and designed a model framework of seven layers for suspicious transaction identification and analysis.Secondly,the Elliptic data set was divided into a training set and test set at a ratio of 7:3.The model was trained and tested with the divided data.The GCN model,Skip-GCN model,EvolveGCN model and other deep neural network models were used as control groups to verify the effectiveness of the model proposed in this paper.Finally,the robustness of the model to the data input was discussed by randomly scrambling the order of each element in the input data.The research results were as follows:The one-dimensional convolutional neural network had good applicability for suspicious transaction recognition.The overall classification accuracy of the Elliptic data set could reach 98%,and the F1value could reach 80%,which had a good classification effect.Comparing the recognition effects of the GCN,Skip-GCN,EvolveGCN and other models on the Elliptic data set,the model proposed in this study had generally good recognition accuracy.The overall accuracy and F1value reached a high level.