Detection of Money Laundering Activities using Supervised Learning Techniques

Abstract

Money laundering is a financial crime that occurs globally, including in Malaysia. The existing traditional detection methods are not adequate to identify complex money laundering activities. This study aims to identify the laundering and non-laundering transactions based on receiving currency, payment currency and payment format using descriptive analysis, apply the supervised learning models, including Random Forest (RF), Naïve Bayes (NB) and K-Nearest Neighbour (KNN), and compare their performances in the classification of laundering and non-laundering transactions. The dataset used in this study was International Business Machines Corporation (IBM) transaction for Anti-Money Laundering (AML) which was available on the Kaggle website with the class imbalance ratio is 0.00077% of laundering cases. The results indicate that the US dollar has the greatest number of laundering and non-laundering transactions in the receiving and payment currency, while the payment format of Automated Clearing House (ACH) has the greatest number of laundering transactions which is 461 equals to 84.1% and the payment format of reinvestment has the greatest number of non-laundering transactions which is 241702 equals to 34.0%. The RF model was not the most effective model in classifying the laundering status of transactions with low MCC value which is 0.0287 and low DOR value which is 41.88. These findings demonstrate that the type of currency and payment format are critical to differentiate the laundering and non-laundering transactions, showing that the supervised learning approach is not an effective predictive model in detecting money laundering activities.