Detection of Illicit Cryptocurrency Transactions Using Hybrid XGBOOST–Random Forest and LIGHTGBM Models

Abstract

Because blockchain networks are decentralized and pseudonymous, money laundering through bitcoin transactions has become a major concern. When dealing with big and complicated transaction datasets, traditional machine learning techniques for unlawful transaction identification frequently have poor generalization and limited accuracy. This paper suggests an improved money laundering detection system that makes use of hybrid machine learning and sophisticated boosting techniques in order to overcome these constraints. By incorporating a hybrid model that combines XGBOOST and Random Forest with potent boosting algorithms like LIGHTGBM and CATBOOST, the system improves detection capabilities above current methods. To find high-frequency and multilayer transaction patterns frequently linked to money laundering operations, Value-driven Transactional Analytics for Crypto Compliance (VTAC) is used. The suggested method greatly enhances predictive performance, as demonstrated by experimental assessment on the Elliptic Bitcoin Dataset, where LIGHTGBM achieved the maximum accuracy of 99.85%. The findings demonstrate that hybrid and advanced boosting models offer better accuracy, scalability, and resilience, which makes the suggested framework ideal for real-time blockchain compliance and anti-money laundering applications.