Analysis of Factors Influencing Fraudulent Transactions in Digital Financial Systems Using Machine Learning Models

Abstrak

This paper explores the use of machine learning, specifically the Random Forest algorithm, to detect fraudulent transactions in digital financial systems. As digital finance grows, the risk of fraud increases, making effective detection systems crucial for maintaining trust and security. The study focuses on identifying key factors influencing fraudulent transactions, such as transaction amount and type, and evaluates the model's performance using accuracy, precision, recall, F1-score, and AUC-ROC metrics. Results show that Random Forest outperforms traditional methods, achieving high accuracy of 95%, precision of 1.00 for fraudulent transactions, and an AUC of 0.98, indicating excellent discriminatory power. By analyzing transaction data, the model identifies important patterns, offering financial institutions practical insights for enhancing fraud detection systems. The findings suggest that focusing on critical features like transaction amount and transfer type can optimize detection systems. However, limitations include the need for further exploration of additional features, such as user behavior, and the integration of more advanced techniques to address emerging fraud tactics. The study’s outcomes provide a robust framework for improving fraud detection in the evolving landscape of digital transactions.