Egonet based anomaly detection in E-bank transaction networks

Abstract

While most of the risks to online banking come from outside, some abnormal accounts can also cause losses to banks or their customers. In this work, we used a data set of 3,411,486 customers collected from a top Chinese bank to mine the abnormal accounts of online banking customers. In order to better express the transaction activities between user accounts, the concept of egonet is used to represent the account relationship in the transaction network. Then, we found that the characteristics of the egonet-based account model conform to the power law distribution, and based on this, anomaly detection is performed. When this method is applied to actual transaction data, we find abnormal account relationships from the data.