Fighting organized crime by automatically detecting money laundering-related financial transactions

Money laundering is the set of operations aimed at giving a legitimate appearance to capital whose origin is illegal, thus making it more difficult to identify and subsequently recover it. It is one of the phenomena on which the so-called underground economy relies and therefore constitutes a crime for which the charge for money laundering applies. For supporting the fight against this phenomenon, the interest towards analysis models for Anti-Money Laundering (AML) based on a combined use of automatic tools and artificial intelligence (AI) techniques increases, as it is also shown by the European Central Bank (ECB) during recent press conferences. Following this direction, this paper proposes a model for enhancing the detection of suspicious transactions related to money laundering. It is based on a set of features that are defined by considering different aspects such as the time, the amount of money, number of transactions, type of operations and level of internationalization. An AI-based computational approach centered on Machine Learning (ML) techniques has been adopted to evaluate the goodness of such feature-based model, in supporting the automatic detection of suspicious transactions, by experimenting 5 different classifiers. From the experiments emerged that the Random Forest provided the best performance not only among the classifiers tested within the paper, but also in comparison to those presented in the related work with an accuracy, a recall and f1-score greater than 94% by decreasing the False Positive Rate (FPR). Furthermore, an analysis on the feature importance has been provided, to understand which feature, among the proposed ones, plays the major role in such application domain.