Improving Classification Performance of Money Laundering Transactions Using Typological Features

The purpose of this research is to analyze the classification performance of the IBM Anti-Money Laundering (AML) dataset with a specific focus on identifying effective and efficient features, including various typologies associated with money laundering: fan-in, fan-out, mutual, and cycle. Additionally, this research conducted a comparative evaluation of these machine learning models, including k-nearest neighbor (KNN), decision tree (DT), support vector machine (SVM), and naïve bayes (NB). The experimental results demonstrated that supervised machine learning models, with the DT leading at 87.6% accuracy, effectively classify money laundering transactions, while other models achieved rates of 87.2% (KNN), 84.4% (SVM), and 71.3% (NB) using the full typology set. Among the examined single typologies, mutual typology consistently outperformed others in terms of accuracy, achieving remarkable rates that surpassed all other single typologies. Specifically, the accuracy rates for different classification models were as follows: 79.3% for KNN, 73.3% for SVM, 76.8% for DT, and 65.2% for NB. These findings underline the significance of using typologies in enhancing the effectiveness of money laundering classification, showcasing its potential as a valuable feature for future anti-money laundering systems.