Prediction of money laundering with machine learning methods using chaotic features

Money laundering poses a significant threat to the security and integrity of the financial systems. Therefore, the effective detection and prevention of money laundering are crucial for the overall security of both financial systems and society. In this study, we aimed to classify transactions in the Bitcoin network as licit or illicit. Using the Elliptic dataset, we transformed the features of the nodes in the graph structure into a time series. We then transferred these time series into phase space and calculated their Lyapunov exponents values. Utilizing feature vectors based on Lyapunov exponents, we performed classification using various machine learning methods. The use of polynomial and interaction features in feature augmentation has notably improved the performance of models by capturing the chaotic structure of the Lyapunov exponents values. The results of the study demonstrated superior performance, with an accuracy of 90.2%, precision of 86.6%, recall of 90.2%, and F1-score of 86.2%.