Applications Of Machine Learning In Bitcoin Anti-Money Laundering

With the popularization and development of cross-border finance and Internet finance,money laundering methods are more complicated and more advanced.At present,more and more money laundering criminals turn their attention from the traditional financial scenario to Bitcoin,which is convenient for money laundering because of its non-sovereignty,disintermediation,anonymity and convenience.Since Bitcoin's entire transaction network data is public,which is different from the confidentiality of data in the traditional financial scenario,it is more necessary to explore a series of anti-money laundering solutions that apply big data technology.In this paper,based on a public Bitcoin transaction dataset from The Elliptic company.several classification models for the licit/illicit are constructed.Supervised classification models are constructed based on the features of transaction entities,using three approaches:Logistic Regression models,Random Forest,and Multilayer Perceptron(MLP).Semi-supervised classification models are constructed based on the features of transaction entities and the transaction flow between transaction entities,using Graph Convolutional Neural Network(GCN).The GCN consists of two layers of graph convolution,which is similar to a perceptron but additionally uses a neighborhood aggregation step motivated by spectral convolution.The improved GCN inserts a skip connection between the intermediate embedding and the input node features.In this paper,the F1 score(the harmonic mean of the precision and the recall)of the illicit category in test set(the harmonic mean of classification accuracy and recall rate)is used to evaluate the models.We have seen Random Forest significantly outperforms other approaches,it also outperforms GCN,even though the latter is empowered by the graph structure information.It reflects the superiority of the tree embedding model in The Elliptic dataset.The GCN outperforms MLP and Logistic Regression,which reflects the improvement of models by adding transaction flow information.This paper shows an idea for constructing an AML monitoring model using GCN based on the graph structure information.