Anomaly Detection Of Blockchain Transaction Data Based On Machine Learning

Originating from Bitcoin,blockchain technology is the underlying core technology of numerous digital currency schemes represented by Bitcoin,which is a decentralized cryptocurrency that allows the secure transfer of funds without a trusted authority.Since the identity information of each node in cryptocurrency does not need to be disclosed or authenticated,information can be transmitted in an anonymous way,which is more likely to generate illegal activities such as money laundering and fraudulent transactions than ordinary currencies.The most important step to deal with the security of transaction data is the exception detection,so it is particularly important to detect the exception of transaction data in blockchain.Based on the Elliptic Data Set,the main contents of this paper are as follows:(1)Aiming at the imbalance problem of blockchain transaction data,this paper uses the Borderline-SMOTE algorithm to process the data set.Unlike the existing oversampling methods,Borderline-SMOTE only oversampling the boundary minority class samples.Optimize the minority class data types by artificial blurring of normal and abnormal data.(2)Aiming at the large dimension of blockchain transaction data,which is easy to cause dimension explosion,this paper solves the problem by dimensionality reduction of data.In this paper,the random forest importance feature selection and auto-encoder importance feature extraction are used to process the data.When the dimension reduction ratio is 0.67,the data dimension reduction effect is optimal,and the data dimension after dimension reduction is 112.According to the SVM classification model and random forest classification model,the data after dimensionality reduction is trained,and the confusion matrix is used to measure the model.It is found that the importance feature extraction of autoencoder can be used to reduce the dimension of the data to better classify the data.(3)Aiming at anomaly detection of blockchain transaction data,this paper adopts a comprehensive learning method,that is,the GRU-SVM model is used to train the data.The SVM classification model and random forest classification model are used to classify the data,and the SVM classification model can better process the data.In this paper,while retaining the original network features of GRUs,SVM classifier is integrated into GRUs network,SVM is used instead of Softmax function to process the data output,and the model is measured by the recall rate and F1 score of evaluation indicators.Compared with the GRU-Softmax model,the F1 score of GRU-SVM model increased by 2.98%,and the recall rate increased by 2%.The new detection method improves the detection level to a certain extent,and can well detect abnormal transaction data,making a certain contribution to the security of blockchain transactions.