Exploration of the effectiveness of expectation maximization algorithm for suspicious transaction detection in anti-money laundering

Abstract

Money laundering refers to activities that disguise money receive through illegal operations and make them become legitimate. It leaves serious consequence that may lead to economy corruption. Extensive research has been conducted to investigate proper solution for suspicious transactions detection. In the realm of clustering approaches, traditional research only concentrate on k-means as the best technique so far. On the other hand, although belongs to the same class, there is a lack of studies conducted in employing Expectation Maximization (EM) for Anti-Money Laundering (AML). The objective of this study is to exploit the advantages of EM for suspicious transaction detection. Data used in this study was obtained through a local bank in Malaysia. Subsets of crucial attributes were selected using genetic search and best first search algorithm. Results indicate that critical fields required for clustering phase include amount, number of credit & debit as well as its sum. The outcome of this study shows that EM overwhelmed traditional clustering method k-means for AML in terms of detecting correct suspicious and normal transactions. This lays the groundwork of employing EM in this field. However, further research is needed using different dataset of other banks in order to clarify the effectiveness of EM in AML.

Variational Autoencoders and Wasserstein Generative Adversarial Networks for Improving the Anti-Money Laundering Process

“…In this paper, we have designed and implemented a deep learning model that gives state-of-the-art results, in terms of the FPR, RFT, and AUC, for improving the anti-money laundering (AML) process. We also explored recent state-ofthe-art deep learning and unsupervised learning techniques such as autoencoder (AE), variational autoencoder (VAE), and generative adversarial network (GAN), and we showed that these techniques can enhance earlier results [7,8].…”

“…The data is obtained from the research project that was undertaken in 2014 between the School of Computer Science, University of Nottingham (Malaysia campus) and a local Malaysian Bank. The original dataset that was obtained in 2014 contains about 30 million transactions (records) for the period from 2012 until 2013 [7]. However, for privacy reasons, the full dataset is not accessible anymore.…”

“…This is because rule-based systems do not consider the economic activities of different people. Furthermore, acquiring economic knowledge about different groups of people can be tedious work on its own [7,8].…”

A Time-Frequency Based Suspicious Activity Detection for Anti-Money Laundering

“…Studies in [11]- [13] applied a cluster-based approach consisting of unsupervised learning techniques such as k-means [14]. Chen et al [15] improved existing outlier detection results with expectation-maximization technique [16]. Unlike the general approach to cluster the customers, Soltani et al [17] cluster the transactions and detect money laundering activities according to structural similarity.…”

Counter Terrorism Finance by Detecting Money Laundering Hidden Networks Using Unsupervised Machine Learning Algorithm

“…Connected Smart Cities 2020; and Web Based Communities and Social Media 2020 criterion for validating clusters after clustering. (Chen et al, 2014) Examined the use of expectation maximization for detecting suspicious financial transaction with 30 million transactions from local X bank in Malaysia. The results show that Expectation Maximization (EM) defeated traditional clustering method k -means for AML in detecting true suspicious transactions with low false positive rate, giving the advantage to EM to be employed in this field.…”