Modelo de segmentación de clientes en el marco del sistema de administración del riesgo de lavado de activos y financiación del terrorismo (SARLAFT) para cooperativas financieras, usando técnicas de Big Data
Loading...
Date
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Católica Luis Amigo.
Abstract
El presente trabajo desarrolla un modelo de segmentación de clientes para una cooperativa financiera en Colombia, utilizando técnicas de Big Data en cumplimiento del Sistema de Administración del Riesgo de Lavado de Activos y Financiación del Terrorismo (SARLAFT). El objetivo central fue clasificar a los clientes según su nivel de riesgo LA/FT mediante la integración de variables normativas, financiera s, sociodemográficas, geográficas y transaccionales. El diseño metodológico se basó en la metodología CRISP - DM, permitiendo estructurar las fases de comprensión del negocio, preparación de datos, modelamiento, evaluación y despliegue del modelo. La base de datos inicial incluyó 43.713 registros de clientes, que fueron depurados y enriquecidos con información de actividad económica, sectores de riesgo y ubicación geográfica, en conjunto con expertos de cumplimiento. Se asignaron ponderaciones y escalas de ri esgo para variables categóricas, y se normalizaron las variables numéricas para garantizar la coherencia analítica. Posteriormente, se implementó un modelo de clustering no supervisado (K - means), terminándose tres clústeres óptimos para personas naturales y jurídicas según las métricas Silhouette Score y Davies - Bouldin. Los resultados evidenciaron clústeres de alto riesgo que requieren debida diligencia ampliada, así como grupos homogéneos de riesgo medio y bajo que permiten focalizar recursos de cumplimien to. El modelo demostró mayor precisión en personas jurídicas, mientras que para personas naturales se identificaron oportunidades de mejora asociadas a calidad de datos. En conclusión, la segmentación propuesta constituye una herramienta eficaz para fortal ecer el cumplimiento normativo SARLAFT, optimizar controles y priorizar esfuerzos de mitigación de riesgos.
This work develops a customer segmentation model for a financial cooperative in Colombia, using Big Data techniques in compliance with the Anti-Money Laundering and Terrorism Financing Risk Management System (SARLAFT). The central objective was to classify clients according to their LA/FT risk level by integrating normative, financial, sociodemographic, geographic, and transactional variables. The methodological design was based on the CRISP-DM methodology, allowing for the structuring of business understanding, data preparation, modeling, evaluation, and model deployment phases. The initial database included 43,713 customer records, which were cleaned and enriched with information on economic activity, risk sectors, and geographic location, in conjunction with compliance experts. Weights and risk scales were assigned for categorical variables, and numerical variables were normalized to ensure analytical coherence. Subsequently, an unsupervised clustering model (K-means) was implemented, resulting in three optimal clusters for natural and legal persons according to Silhouette Score and Davies-Bouldin metrics. The results showed high-risk clusters requiring enhanced due diligence, as well as homogeneous groups of medium and low risk that allow for the targeting of compliance resources. The model demonstrated greater precision for legal entities, while for natural persons, improvement opportunities associated with data quality were identified. In conclusion, the proposed segmentation constitutes an effective tool to strengthen SARLAFT regulatory compliance, optimize controls, and prioritize risk mitigation efforts.
This work develops a customer segmentation model for a financial cooperative in Colombia, using Big Data techniques in compliance with the Anti-Money Laundering and Terrorism Financing Risk Management System (SARLAFT). The central objective was to classify clients according to their LA/FT risk level by integrating normative, financial, sociodemographic, geographic, and transactional variables. The methodological design was based on the CRISP-DM methodology, allowing for the structuring of business understanding, data preparation, modeling, evaluation, and model deployment phases. The initial database included 43,713 customer records, which were cleaned and enriched with information on economic activity, risk sectors, and geographic location, in conjunction with compliance experts. Weights and risk scales were assigned for categorical variables, and numerical variables were normalized to ensure analytical coherence. Subsequently, an unsupervised clustering model (K-means) was implemented, resulting in three optimal clusters for natural and legal persons according to Silhouette Score and Davies-Bouldin metrics. The results showed high-risk clusters requiring enhanced due diligence, as well as homogeneous groups of medium and low risk that allow for the targeting of compliance resources. The model demonstrated greater precision for legal entities, while for natural persons, improvement opportunities associated with data quality were identified. In conclusion, the proposed segmentation constitutes an effective tool to strengthen SARLAFT regulatory compliance, optimize controls, and prioritize risk mitigation efforts.
Description
Citation
Villa Álvarez, J. D., Valencia Restrepo, D., & Arbeláez Zapata, C. A. (2025). Modelo de segmentación de clientes en el marco del sistema de administración del riesgo de lavado de activos y financiación del terrorismo (SARLAFT) para cooperativas financieras, usando técnicas de Big Data