Abstract: Aimed at the problem that anti-money laundering monitoring in financial system relies too much on computer screening, which leads to a low accuracy and adaptability, a man-machine separation mode of anti-money laundering monitoring system is improved to the man-machine coupling mode, and the machine learning algorithm is trained with real transaction data to optimize the model. After validation and evaluation outside the sample, it is found that the machine learning model based on the random forest algorithm has a better accuracy and applicability in the man-machine coupling monitoring system. The monitoring system has a stronger adaptability to the new money laundering mode. While ensuring the accuracy of monitoring, it reduces the number of false early warning of suspicious transactions, and improves the efficiency of payment institutions in monitoring suspicious money laundering transactions.

Key words: machine learning, random forest algorithm, anti-money laundering, profiling system

摘要： 针对金融系统反洗钱监测过于依赖计算机筛查导致准确率和适应性较低的问题，将反洗钱监测系统的人机分离模式改进为人机耦合模式，并用真实交易数据训练机器学习算法以优化模型。经样本外验证和评估后发现，在人机耦合监测系统下，基于随机森林算法的机器学习模型具有更好的准确性和适用性。该监测系统对新型洗钱方式具有更强的适应能力，在保证监测准确率的同时，降低了可疑交易错误预警次数，提高了支付机构对可疑洗钱交易监测的效率。

关键词: 机器学习算法, 随机森林算法, 反洗钱, 监测系统