Machine Learning and Sampling Scheme: An Empirical Study of Money Laundering Detection

Abstract

This paper studies the interplay of machine learning and sampling scheme in an empirical analysis of money laundering detection algorithms. Using actual transaction data provided by a U.S. financial institution, we study five major machine learning algorithms including Bayes logistic regression, decision tree, random forest, support vector machine, and artificial neural network. As the incidence of money laundering events is rare, we apply and compare two sampling techniques that increase the relative presence of the events. Our analysis reveals potential advantages of machine learning algorithms in modeling money laundering events. This paper provides insights into the use of machine learning and sampling schemes in money laundering detection specifically, and classification of rare events in general.