Research On Illegal Transaction Detection Of Blockchain Based On Machine Learning

As one of the most valuable cryptocurrencies and platforms in blockchain technology,the security of transactions based on Bitcoin and Ethereum has also become a research hotspot in industry and academia.Although both Bitcoin and Ethereum are based on blockchain technology,there are different.Bitcoin is a cryptocurrency based on blockchain.Bitcoin transactions are based on a decentralized mechanism that lacks centralized control and is difficult to trace.Therefore,criminals are keen to use Bitcoin to achieve low-cost,peer-to-peer cross-border transactions.Ethereum is an open-source blockchain distributed platform.Due to the lack of regulation and anonymity of Ethereum,some speculators use this feature to conduct money laundering,financial fraud,phishing,Ponzi scheme,and other criminal activities on Ethereum.Therefore,this paper aims to comprehensively apply machine learning technology to study the detection methods of common illegal transactions in various accounts based on Bitcoin and Ethereum transaction data,to promote the sustainable development of the blockchain ecosystem.The main work of this study is as follows:(1)This paper proposes a method for detecting bitcoin illegal transactions based on representation learning multi-scale Graph Convolutional Neural Networks.In this paper,the Elliptic data set published by Elliptic is used.In the first step,the graph structure is optimized by using self-supervised presentation learning to enhance the robustness of the model.Firstly,the disturbance graph is preprocessed roughly.Then,trim the edges whose scores are below the threshold;Finally,the features of the nodes in the preprocessed graph are randomly scrambled and recovered.The second step is to enhance the graph after obtaining a high-quality representation.The third step is to input the refined reconstructed graph into the downstream classifier and use the multi-scale graph convolutional neural network(M-GCN)for classification.This classifier can aggregate the features of adjacent nodes to make better use of the connections between different Bitcoin transactions and improve the accuracy of detection.Experimental results show that the detection effect of this paper is improved compared with previous studies.(2)A multi-view-based Ethereum illegal transaction detection method is proposed.This paper uses 2,179 illegal transaction accounts identified by the My Crypto team and the Ethereum community and 7,662 normal accounts disclosed by Farrugia et al.This paper uses Bi-directional Long Short-term Memory(Bi LSTM)and One-dimensional Convolutional Neural Networks(1D CNN)to share opcode view features,bytecode view features,and account transaction view features among all views to learn important semantic information between views.In addition,this paper adds the contrast learning method to train multi-view features so that the correlation between view features is stronger.As a result,potential semantic information can be captured more accurately,and illegal transactions can be predicted more accurately.Experimental results show that the detection effect of this paper is improved compared with previous studies.