Machine Learning-Based Pattern Recognition for Anti-Money Laundering in Banking Systems

Abstract

This paper presents a new machine-based awareness system for anti-money laundering (AML) in banks. The planning process combines data pre-processing techniques, engineering techniques, and integrated models to address the limitations of legal compliance under AML. The methodology incorporates a multi-layered approach combining supervised and unsupervised learning components to enhance detection accuracy while maintaining computational efficiency. Experimental evaluation was conducted using a comprehensive dataset comprising over 2 million financial transactions spanning 24 months from multiple banking institutions. The system demonstrates significant improvements in detection capabilities, achieving a 22.3% increase in recall while maintaining false positive rates below 3%. The implementation of adaptive threshold mechanisms and dynamic feature selection techniques contributes to a 25.2% improvement in AUC-ROC scores compared to conventional methods. The system's scalability has been validated through extensive testing, maintaining detection accuracy under high-load conditions processing up to 15,000 transactions per second. The research establishes new benchmarks for AML system performance and provides empirical evidence supporting the effectiveness of machine learning techniques in operational environments. The processing process has a way for the AML performance and presenting a great deal of research in the future research.