Simulation Of Bank Transaction Data

When working with data such as financial transactions or user activity logs, in domains with inherent privacy concerns, you will certainly run into problems with data protection and data availability. Among possible approaches to cope with these problems are data anonymization and data simulation. One of essential advantages in favor of data simulation is complete separation from subject identification.In this paper, we present a multi-agent simulation framework applicable in financial domain for transaction data generation. We use the framework to create multiple simulation scenarios reproducing typical fraudulent behavioral patterns and we evaluate its applicability on the task of Anti Money Laundering (AML) by classification of fraudulent agents.