A Machine Learning Approach to Analyze and Support Anti-Corruption Policy

Abstract

Can machine learning support better governance? This study uses a tree-based gradient-boosted classifier to predict corruption in Brazilian municipalities using budget data as predictors. The trained model offers a predictive measure of corruption, which we validate through replication and extension of previous corruption studies. Our policy simulations show that machine learning can significantly enhance corruption detection: compared to random audits, a machine-guided targeted policy could detect almost twice as many corrupt municipalities for the same audit rate.

Keywords: algorithmic decision-making, corruption policy, local public finance

JEL Classification: D73, E62, K14, K42