块链
计算机科学
数据库事务
数据挖掘
特征(语言学)
人工智能
机器学习
数据库
计算机安全
哲学
语言学
作者
Hualong Han,Ranran Wang,Yvpeng Chen,Kai Xie,Ke Zhang
出处
期刊:Communications in computer and information science
日期:2022-01-01
卷期号:: 223-236
被引量:2
标识
DOI:10.1007/978-981-19-8043-5_16
摘要
Blockchain has become a hot topic in current academic research due to its decentralization, imtamability and anonymity. However it also makes blockchain a tool for money laundering, fraud, extortion and other illegal activities. Therefore, it is particularly important to supervise and detect abnormal transactions on blockchain. In our paper, graph data structure is used to express blockchain node transactions. Aiming at the problem of too few abnormal transaction samples of block chain, k-rate sampling and feature similarity are used to solve the problem of unbalanced data. Further more, we select some features for feature preprocessing according to the distribution of multiple features of blockchain transactions. Finally, the blockchain transaction topology is used for multi-graph convolutional neural network machine learning methods to train the blockchain transaction data. So the abnormal nodes and non-abnormal nodes classification model is obtained. And the abnormal transactions of blockchain is detected by using the classification model. Experimental results show that compared with traditional algorithms such as logistic regression (LR), multi-layer perceptron (MLP) and linear regression (LR), the anomaly detection algorithm based on sample equalization and feature engineering has significantly improved recall rate. This provides a theoretical and practical basis for our next work – tracing abnormal nodes of blockchain, which has a good prospect for industrial application.
科研通智能强力驱动
Strongly Powered by AbleSci AI