MUSEM: combining multi-up sampling and ensemble learning methods for effective financial fraud detection

The rise of electronic payments, both online and in-person, has coincided with an increase in fraudulent and defaulted transactions, leading to significant financial losses. Researchers have explored various machine learning models for anomaly detection in credit card transactions, but challenges such as overlapping data classes and imbalanced distributions persist. To address these issues, we propose a dual-strategy approach called MUSEM, which integrates multi-up sampling with ensemble learning for enhanced fraud detection. MUSEM combines seven individual models into a unified framework, offering a more efficient method for identifying fraud. This study presents a comprehensive review and comparative analysis of various machine learning algorithms employed in financial fraud detection. Experimental results demonstrate a 3% improvement in recall over individual classifiers, affirming the effectiveness of the ensemble learning paradigm adopted in MUSEM. The findings highlight MUSEM's potential for real-world fraud detection applications, improving electronic payment security and reducing financial risks.