Modeling Digital Repression: A Machine Learning Analysis of Shutdowns as Governance Signals

Abstract

This study advances Digital Government research by applying machine learning to analyze internet shutdowns as structured signals of digital repression. Using a dataset of 566 shutdown events (1995–2011) and regime attributes from the Polity 5 project, the study builds interpretable models to estimate shutdown severity and classify regime type. A decision tree regressor and bootstrapped logistic classifier reveal strong associations between shutdown characteristics and political context, achieving over 93% accuracy. While not designed for real-time prediction, these models demonstrate how event-level data can inform early warning, policy evaluation, and digital rights monitoring. By modeling shutdowns as governance decisions embedded in digital infrastructure, this research shows how computational methods can support accountability in opaque information environments.