Generación de datos sintéticos orientada a la detección del la/ft mediante simulación basada en agentes

Abstract

The central challenge in detecting money laundering and terrorist financing lies in the limited availability of real transactional data for developing and validating advanced analytics models. Restrictions stemming from personal data protection and banking secrecy prevent the scientific community from accessing the information needed to train, compare, and reproduce research, creating a reliance on pre-existing or private synthetic datasets that cannot be shared. This situation hinders the objective evaluation of algorithms—particularly those based on unsupervised learning—and limits the simulation of new criminal typologies not yet reflected in historical records. Consequently, methodological innovation and the development of adaptable solutions capable of addressing the constantly evolving strategies of criminal organizations are constrained. In this context, it is essential to develop a methodological framework that enables the generation of synthetic data from scratch, grounded in expert knowledge rather than the replication of real-world information. Integrating fuzzy systems and agent-based simulation offers a promising alternative for modeling the uncertainty inherent in human behavior and replicating financial system dynamics with a high degree of realism, all while preserving privacy and fostering experimentation, scientific reproducibility, and enhanced capabilities for detecting illicit activities