关键词: 数字货币, SHAP, XGBoost, mRMR, 集成学习, Elliptic数据集

Abstract: The popularity of digital currencies has contributed to the increasing scale of the trading market, which is also accompanied by the aggravation of the problem of abnormal behavior in trading. The existing anomaly recognition methods have shortcomings such as relying on complex rules, lack of flexibility and adaptability, and are difficult to process large-scale data efficiently. Therefore, the study proposes a machine learning-based abnormal behavior identification model for digital currency trading. The dataset features are analyzed by using the Shapley value additivity interpretation analysis, and the recognition model is constructed by combining the minimum redundancy maximum correlation algorithm and extreme gradient boosting algorithm. Experiments show that the proposed model takes 11.13%, 19.13%, and 16.37% less time than the other 3 models. The average F1 score of the model in the three data tables is 0.920, which is an average increase of 6.33% over the other three methods. The results show the superiority and effectiveness of the identification model proposed in the study in the field of digital currency transaction anomaly identification. The application of Shapley value additivity interpretive analysis to the identification of trading anomalies provides new perspectives on the model's interpretability and transparency. The anomaly identification model is of great significance in preventing and identifying anomalous behaviors in digital currency transactions.

Key words: Digital currency, SHAP, XGBoost, mRMR, Integrated learning, Elliptic datase