The best machine learning model for fraud detection on e-platforms: a systematic literature review

Abstract

The internet has been instrumental in the development and facilitation of online payment systems. However, its associated fraudulent activities on e-platforms cannot be overlooked. As a result, there has been a growing interest in the application of machine learning (ML) algorithms for fraud detection on financial e-platforms. The goal of this research is to identify common types of fraud on financial e-platform, highlight different machine learning algorithms employed in fraud detection, and derive the best machine learning algorithms for fraud detection on e-platforms. To achieve this goal, the research followed a nine steps systematic review approach to retrieve Journals and conference publications from science direct, Google Scholar and IEEE Xplore between 2018 and 2023. Out of 2,071 articles identified and screened, 44 publications (23 articles and 21 conference proceedings) satisfied the inclusion criteria for further analysis. The random forest algorithm turned out to be the best ML algorithm because it ranked first in the frequency of usage analysis and ranked first in the performance analysis with an average accuracy of 96.67%. Overall, this review has identified the kinds of fraud on financial e-platforms, and proclaimed the best and least ML algorithm for fraud detection on financial e-platform. This can help guide future research and inform the development of more effective fraud detection systems.