AI for integrity: Predicting public servants' susceptibility to corruption via supervised machine learning

Public corruption undermines government performance, erodes trust, and fuels illiberal populism. While macro-level drivers of corruption are well documented, little is known about the individual-level attitudes and beliefs associated with public servants' susceptibility to engaging in corrupt behaviour. Using a large-scale, globally stratified dataset of 18,277 public servants across 90 countries, we systematically compare a conventional regression baseline with five contemporary supervised machine learning models to predict susceptibility to corruption. Machine learning improves predictive accuracy and shows that democratic values, beliefs about competition, and attitudes towards leadership are more consistently predictive than socio-economic characteristics such as income, education, or gender. These patterns are robust across modelling approaches and analytical settings. The findings illustrate the value of transparent model comparison and machine learning for studying integrity risks, provide exploratory insights for corruption prevention strategies, and open new avenues for research on predictive modelling for public sector integrity.