Detecting Pyramid Scheme Accounts with Time Series Financial Transactions

Abstract

The traces of pyramid scheme crimes are exposed to the floating of capital data. Existing studies on the time series financial data have primarily focused on identifing accounts' trading patterns. In this paper, however, we creatively leverage such data for identifying pyramid scheme accounts in the complex financial network. Indeed, detecting pyramid scheme account through the financial data is critical for improving customer satisfaction and financial security. However, in practice, it is challenging to identifying the anomalous accounts from regular ones. To this end, on the basic analysis of transaction records, we propose an account behavioral profiling framework and enter the obtained features to detector. Specifically, we first classify the role-based responsibility of multi-layer marketing members. Then we extracted a number of behavioral features from each account's sequential transaction records. Then we fed outlier detection methods with our profiling to identify pyramid scheme accounts, who exhibit anomalous transaction behaviors. Experimental results demonstrated the effectiveness of our profiling features.