Weak Supervision Can Help Detecting Corruption in Public Procurement

Corruption in public procurement in the European Union is estimated to cost more than 120 billion Euros per year. Detecting corruption in public procurement is challenging because of imperfect corruption reg flags and vast number of tenders. Authorities need potent and scalable tools to identify corrupt public procurement practices. This paper shows that weak supervision machine learning algorithm provides a proficient method that can easily be implemented. Authorities can effectively calculate corruption probabilities of millions of public procurement tenders using the weak supervision algorithm. Weak supervision combines information contents of imperfect corruption red flags with unknown accuracies. Additionally, it can handle measurement errors. I analyze potential corruption in 25,859,734 European Union (EU) public procurement contracts in years 2009–2020 using the Snorkel weak supervision algorithm. These contracts are awarded by 1,212,533 authorities in 33 EU and affiliated countries. The analysis suggests that 40% of contracts and 22% of EU authorities are susceptible to corruption. Experimental results show that training machine learning models with weak supervision labeled data produces superior prediction results.