GraphSIF: analyzing flow of payments in a Business-to-Business network to detect supplier impersonation

Abstract

Supplier Impersonation Fraud (SIF) is a rising issue for Business-to-Business companies. The use of remote and quick digital transactions has made the task of identifying fraudsters more difficult. In this paper, we propose a data-driven fraud detection system whose goal is to provide an accurate estimation of financial transaction legitimacy by using the knowledge contained in the network of transactions created by the interaction of a company with its suppliers. We consider the real dataset collected by SIS-ID for this work.We propose to use a graph-based approach to design an Anomaly Detection System (ADS) based on a Self-Organizing Map (SOM) allowing us to label a suspicious transaction as either legitimate or fraudulent based on its similarity with frequently occurring transactions for a given company. Experiments demonstrate that our approach shows high consistency with expert knowledge on a real-life dataset, while performing faster than the expert system.