Application of Gradient Boosting Algorithms for Anti-money Laundering in Cryptocurrencies

摘要: The recent emergence of cryptocurrencies has added another layer complexity in the fight towards financial crime. Cryptocurrencies require no central authority and offer pseudo-anonymity to its users, allowing criminals disguise themselves among legitimate users. On other hand, openness data fuels investigator’s toolkit conduct forensic examinations. This study focuses on detection illicit activities (e.g., scams, financing terrorism, Ponzi schemes) cryptocurrency infrastructures, both at an account transaction level. Previous work identified that class imbalance dynamic environment created by evolving techniques deployed avoid are widespread this domain. In our study, we propose Adaptive Stacked eXtreme Gradient Boosting (ASXGB), adaptation (XGBoost), better handle environments present a comparative analysis various offline decision tree-based ensembles heuristic-based data-sampling techniques. Our results show that: (i) gradient boosting algorithms outperform state-of-the-art Random Forest (RF) level, (ii) approach NCL-SMOTE further improves recall (iii) proposed ASXGB successfully reduced impact concept drift while improving