Research On The Identification Of Abnormal Transaction Behavior In Ethereum Based On Machine Learning

At present,our country is actively promoting the integrated development of blockchain technology and economic society.However,as a digital economy infrastructure,digital currency has the characteristics of protecting privacy,making it a "hotbed" for criminals to carry out illegal activities.This phenomenon has brought challenges for our country to promote the development of the digital economy and create a safe,stable and healthy financial environment.Therefore,identifying the security risks caused by abnormal transactions from the massive transaction data,forming a regulatory system,and promoting the healthy development of the digital economy has become one of the important challenges of digital currency monitoring and supervision.All digital currency transaction ledgers are publicly accessible.At present,the mainstream method for identifying abnormal trading behaviors of digital currencies is to use capital flow analysis methods or machine learning methods.However,these methods have large limitations,poor timeliness,and recognition accuracy that needs to be improved.Therefore,this paper selects the Ethereum digital currency as the research object,and proposes an efficient,automatic,and accurate method for identifying abnormal transaction behaviors that can solve the problems of large limitations and poor timeliness.The research problem of this paper is essentially a classification problem,so this paper adopts machine learning classification method to realize the identification of abnormal transaction behaviors in Ethereum.The main work of this paper is as follows:(1)In response to the problem of data collection,first deploy the Ethereum full node of Blockchain Data Service to achieve real-time acquisition of the Ethereum transaction ledger;then use a third-party website to collect transaction accounts whose transaction behavior has been verified.(2)A recognition system based on traditional machine learning is proposed.In the feature engineering aspect of the system,for the problem of how to select effective features to improve the performance of the model obtained by training,first,according to all transaction records,the feature extraction of transaction attributes is performed for each sample account to form a sample data set,and all samples Set the input to the ET model to obtain the importance value of each feature,thereby determining the selection plan of the basic feature.In terms of classifiers,this paper selects logistic regression,decision trees and naive Bayes classification models as the classifiers in the system.(3)A recognition system based on graph neural network is proposed.In this proposed system,first,a feature extraction method based on network embedding is proposed for the need to consider both the structural characteristics of the transaction network and the physical signs of transaction attributes.Then in terms of the classifier,the graph convolutional neural network model was selected to improve the accuracy of the classification model.The final experimental results show that the Ethereum abnormal transaction behavior identification system based on graph neural network proposed in this paper can reach 95% accuracy,which reflects the excellent performance of the system in the identification of fraudulent transactions in Ethereum.