Abstract:
Money laundering remains a critical global threat, with financial institutions struggling to
detect complex and evolving laundering schemes within vast volumes of transactional data.
This study focuses on identifying suspicious money laundering behaviors by employing
machine learning (ML) techniques, with a focus on improving detection accuracy while
ensuring interpretability for compliance and auditability.
To address this issue, a supervised ML-based AML detection system was developed using the publicly available IBM Transactions for Anti-Money Laundering (AML) dataset. The research methodology included extensive data pre-processing, feature engineering, and the evaluation of various classification models such as XGBoost, LightGBM, CatBoost, Random Forest, and Logistic Regression. Advanced techniques like ensemble learning and hyperparameter tuning were applied, and model performance was evaluated using standard metrics including F1 Score, PR AUC, and MCC.
The research revealed that the weighted ensemble model combining XGBoost, LightGBM,
and CatBoost delivered the best performance, achieving an F1 Score of 0.9923, PR AUC of
0.9717, and MCC of 0.8444. These results indicate the system’s strong potential to support
AML efforts by identifying suspicious transaction patterns with higher accuracy and reliability than traditional models.
