Performance Analysis of Machine Learning Techniques for Detecting Money Laundering in Bitcoin Transactions

Abstract

More people are relying on Cryptocurrencies, yet these currencies are not controlled by any central authority or governments. Bitcoin transactions' anonymity facilitates money laundry activities by cyber-criminals. Traditional Anti-Money Laundering solutions, often rule-based, suffer from high false positive rates, leading to substantial operational costs. Machine Learning (ML) algorithms offers efficient analysis and identification of abnormal patterns in vast amounts of data; thus predicts suspicious transactions. This paper evaluates the effectiveness of four different ML algorithms in classifying Bitcoin transactions as either licit, or illicit transactions. The ML models were trained on the Elliptic dataset of Bitcoin transactions categorized into real entities belonging to licit and illicit categories [1]. The models were evaluated by the confusion matrix, accuracy, and features importance study. The experimental results proved that Random Forest (RF) and Extreme Gradient Booster (XGBoost) models achieved the highest accuracy of 95.1 % and 95.6%, respectively. This indicates that RF and XGBoost models effectively managed the data's complexity and accurately predicted suspicious transactions.