Detecting Money Laundering Patterns through Cash Flow Analysis: a Neural Network-Based Approach

Money laundering poses a serious threat to the international financial system and requires effective detection methods. The techniques used by reporting entities in the AML/CFT regime may be incapable of identifying advanced financial frauds that have a convoluted nature as well as of processing large amounts of data. This paper builds an ANN based on analyzing cash flows in bank statements in order to identify patterns of money laundering. The toolkit is automated using Python programming language in Jupyter Notebook. The efficiency of the ANN is evaluated using ROC curves and a confusion matrix. In addition, the accuracy of the ANN is compared with that of other common machine learning models, including decision trees, random forest and gradient boosting. The results demonstrate the high efficiency of the ANN, which indicates the rationality and feasibility of the neural network-based approach to cash flow analysis for the purpose of identifying money laundering patterns. The toolkit is recommended for adoption in the operational processes of reporting entities to improve the quality of internal control for AML/CFT purposes.