Blockchain Abnormal Transaction Detection Based on Feature Selection and Ensemble Algorithm
Because of its open and immutable characteristics,blockchain provides convenience for virtual currency transactions,and at the same time,virtual currency has become an important carrier of criminal acts such as money laundering and theft,which brings huge risks to citizens'property security.There-fore,it is of great significance to detect the abnormal transactions of blockchain efficiently and quickly.The anomaly detection model based on full features is often interfered by redundant features,which af-fects the efficiency and accuracy of detection.To solve these problems,a blockchain anomaly detection method based on a two-layer feature selection and ensemble algorithm was proposed.Firstly,the features with low correlation with the target variable were removed by Chi-square test,and the first layer feature set was obtained.Secondly,some highly correlated features were removed by calculating the Pearson cor-relation coefficient between the features,and the second layer feature set was obtained.Finally,a block-chain abnormal transaction detection model based on feature selection and ensemble algorithm was estab-lished.The results show that the two-layer feature selection model is superior to the all-feature ensemble learning model and other well-known anomaly detection models,and the precision,recall and F1 score of the proposed feature selection model are improved by 2.2％,0.15％and 1.0％on average,respective-ly,and the training time is saved by about 20％.
万方数据
