Detection of fraud in public procurement using data-driven methods: a systematic mapping study

Abstract The scientific literature dedicated to the detection of fraud in public procurement is vast, with several studies reporting the use of different methodologies to detect corruption. However, the literature still lacks a comprehensive study of the types of fraud being investigated and how data-driven techniques are being used to address this problem. This article aims to provide a better overview of how these techniques are used to detect corruption in public procurement. We systematically searched academic databases with the goal of finding papers that used data-driven techniques to predict or identify fraud in public procurement. We also performed a snowballing procedure to complement the database search with additional papers. 93 works were added to our study after screening and evaluation of more than 6000 papers. Relevant information was extracted from these papers to answer the research question defined during the planning phase. The results showed that most works use machine learning models to detect collusion and statistical analysis to detect instances of favoritism. Despite the promising results, there are some gaps that still need to be addressed. There is a lack of papers that employ the proposed methodologies in real-life systems to detect new cases of corruption. Another gap found is the lack of public available datasets, hindering the replication and dissemination of the proposed methodologies. The findings of our study contribute to a more comprehensive understanding of fraud detection in public procurement, pointing to areas for improvement and offering insights to researchers and institutions seeking to improve their processes.