KOEX

BMBF funded project: Collaborative Machine Learning for the Detection of Fraud and Risks in ERP Systems (KOEX)

Motivation and project goals

The widespread use of IT systems, and ERP systems in particular, for the management of corporate processes has opened up a wide range of new points of attack for fraudulent or erroneous activities. While the focus of current security solutions is usually on external attacks, internal misuse, error, incorrect operation and fraud are often considered of secondary importance. The Association of Certified Fraud Examiners estimates the damage caused to companies by fraud at around 5% of their annual turnover^[1].

Current ERP systems, most notably the SAP SE system, manage large amounts of data that provide information about the activities of the employees as well as logging the entire flow of goods and finances of the company. Traditional methods mostly use predefined rules to detect fraud in this data. However, new fraud cases are not detected by these approaches.

The aim of the KOEX project is to automatically identify known and unknown fraud cases in the data using machine learning methods. Identified fraud patterns are to be abstracted and made usable for other companies using federated learning techniques. Thus, no confidential details and personal data will be released. The findings from individual fraud cases can therefore be used directly without revealing the source. The diverse competencies of the partners involved will be used to develop an initial demonstration system within the framework of this project, which will be gradually extended and incorporated into the existing SIVIS suite of tools at the end of the project.

[1] ACFE. Report to the Nations (2020): acfepublic.s3-us-west-2.amazonaws.com/2020-Report-to-the-Nations.pdf