Automated refund fraud detection in mobile money transactions using Machine Learning

Abstract

In an era marked by the widespread adoption of mobile money systems, the need to secure financial transactions against fraudulent activities has never been more pressing. This research addresses the crucial issue of refund fraud detection within the realm of mobile money transactions, employing machine learning techniques as a formidable weapon against this evolving threat. The primary objective of this research is to develop a machine-learning model for detecting refund fraud in mobile money transactions, with a specific focus on evaluating the efficacy of machine learning models, including Naive Bayes, Logistic Regression, and XG Boost. Our research encompasses a comprehensive methodology that includes data collection, preprocessing, feature engineering, model selection, training, and evaluation. Leveraging a dataset comprising legitimate and fraudulent mobile money transactions, we meticulously prepared the data, engineered features, and rigorously evaluated the performance of these models. Results from our experiments revealed that the XG Boost algorithm emerged as the most e↵ective model for detecting refund fraud in mobile money transactions. With an exceptional F1-Score of 0.828224, XG Boost demonstrated a remarkable balance between precision and recall, highlighting its capability to distinguish between legitimate and fraudulent transactions, and significantly contribute to enhanced security in mobile money systems. This research underscores the vital role of machine learning, particularly the XG Boost algorithm, in fortifying the security of mobile money transactions by automating the detection of refund fraud. Our findings not only advance the understanding of fraud detection in the realm of financial technology but also provide actionable insights for industry stakeholders and policymakers, paving the way for more secure and trustworthy mobile money ecosystems.