Money Laundering Detection in Financial Institutions Using Machine Learning

Abstract

This research aims to develop a system based on machine learning techniques for the accurate detection and classification of money laundering transactions. This system utilizes real and synthetic financial data, including the IBM AML dataset and synthetic financial databases. The methodology included data collection and cleaning, feature extraction from transaction behavior, and training several classification algorithms, including Random Forest, Support Vector Machine (SVM), and LogitBoost. A hybrid model was also built using a stacking approach to leverage the advantages of each algorithm. The models were evaluated using various statistical measures, including precision, recall, and the F1 score, as well as performance analysis via ROC and PR curves. The results demonstrated that the hybrid model outperformed individual models in detecting suspicious transactions and reducing errors, enhancing its effectiveness as an intelligent tool to support financial institutions and regulatory authorities in the early detection of financial crimes. The research recommends expanding the scope of data used and leveraging deep learning techniques to enhance the system's efficiency and accuracy.