Money laundering is the process of putting the money obtained from illegal transactions into legal financial systems, which is often achieved by many transfers to make the illegal origins hard to be traced. More specifically, money laundering could be divided into three stages including placement which is the first time putting illegal money into financial institutions, layering which is splitting the money into various bank accounts to hide its illegal sources, and integration which is collecting back the clean/white money in the end.
Anti-money laundering is a series of regulations that financial institutions should follow to prevent suspicious transactions. Anti-money laundering has received high attention in most countries since money laundering usually involves criminal or terrorist activities. Banks over the world are working hard to implement anti-money laundering to prevent becoming a money laundering haven that would cause huge damage to social and financial stability. Banks with impaired reputations would even hinder international cash flow transfers and be isolated by other countries.
Traditionally, anti-money laundering compliance is mainly done by human work, which is a significant burden to banks. In recent times, more and more research has proven the potential of machine learning in anti-money laundering (Chen et al., 2018). Supervised machine learning techniques such as support vector machine, decision tree, and k-nearest neighbor can be applied to detect suspicious transactions (Tang and Yin,2005; Wang and Yang, 2007; Khodabakhshi and Fartash, 2016). Machine learning models utilize labeled historical data from bank internal systems to classify the risk of accounts and can be taught to keep fine-tuning the models and then deliver optimal classification.
With the help of machine learning, the effectiveness of anti-money laundering compliance could be increased by reducing human errors and relieving workers from repeated paperwork to tasks with higher value. However, some concerns about using machine learning in anti-money laundering still remained. Since banks performed quite well in the past anti-money laundering activities, they may have little data labeled as money laundering, so the historical data might be extremely imbalanced. Besides, if banks want to increase the accuracy of alerts predicted from machine learning models, they may need to collect more information from customers which may put negative impacts on customer experience due to personal privacy protection. Therefore, banks still have to evaluate some tradeoffs when they are going to replace human work with emerging technologies.
References:
Javatpoint, Anti-Money Laundering using Machine Learning
https://www.javatpoint.com/anti-money-laundering-using-machine-learning
Financial Action Task Force (FATF), Money Laundering
https://www.fatf-gafi.org/faq/moneylaundering/
Bearingpoint, How machine learning can dramatically reduce financial institutions’ cost of compliance
https://www.bearingpoint.com/en-no/insights-events/insights/machine-learning-is-the-key-to-efficient-and-effective-aml/
Chen, Z., Van Khoa, L. D., Teoh, E. N., Nazir, A., Karuppiah, E. K., & Lam, K. S. (2018). Machine learning techniques for anti-money laundering (AML) solutions in suspicious transaction detection: a review. Knowledge and Information Systems, 57(2), 245-285.
Tang, J., & Yin, J. (2005, August). Developing an intelligent data discriminating system of anti-money laundering based on SVM. In 2005 International conference on machine learning and cybernetics (Vol. 6, pp. 3453-3457). IEEE.
Wang, S. N., & Yang, J. G. (2007, August). A money laundering risk evaluation method based on decision tree. In 2007 International conference on machine learning and cybernetics (Vol. 1, pp. 283-286). IEEE.
Khodabakhshi, M., & Fartash, M. (2016, November). Fraud detection in banking using knn (k-nearest neighbor) algorithm. In International Conf. on Research in Science and Technology.
