Abstract
This study develops a Global Early-Warning System (GEWS) using machine learning to predict political instability events from integrated global datasets. Data were compiled from the Armed Conflict Location & Event Data Project (ACLED), World Bank indicators, the Varieties of Democracy (V-Dem) dataset, and the Global Database of Events, Language, and Tone (GDELT), covering 2000–2023. The final dataset included 265 countries and territories, 5,830 country-year observations, and 662 instability events (11.36% prevalence). Five models were evaluated: Logistic Regression, Random Forest, Gradient Boosting, XGBoost, and Long Short-Term Memory (LSTM). Ensemble models performed best, with Random Forest achieving the highest accuracy (AUC = 0.95), followed by XGBoost (AUC = 0.94). Key predictors included population size, GDP per capita, governance indicators, and inflation volatility. The model generates a Political Instability Risk Score and can identify risks 6–12 months in advance.
